]> git.ipfire.org Git - thirdparty/linux.git/blame - fs/btrfs/block-group.c
btrfs: handle space_info setting of bg in btrfs_add_bg_to_space_info
[thirdparty/linux.git] / fs / btrfs / block-group.c
CommitLineData
2e405ad8
JB
1// SPDX-License-Identifier: GPL-2.0
2
2ca0ec77 3#include <linux/list_sort.h>
784352fe 4#include "misc.h"
2e405ad8
JB
5#include "ctree.h"
6#include "block-group.h"
3eeb3226 7#include "space-info.h"
9f21246d
JB
8#include "disk-io.h"
9#include "free-space-cache.h"
10#include "free-space-tree.h"
e3e0520b
JB
11#include "volumes.h"
12#include "transaction.h"
13#include "ref-verify.h"
4358d963
JB
14#include "sysfs.h"
15#include "tree-log.h"
77745c05 16#include "delalloc-space.h"
b0643e59 17#include "discard.h"
96a14336 18#include "raid56.h"
08e11a3d 19#include "zoned.h"
2e405ad8 20
878d7b67
JB
21/*
22 * Return target flags in extended format or 0 if restripe for this chunk_type
23 * is not in progress
24 *
25 * Should be called with balance_lock held
26 */
e11c0406 27static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
878d7b67
JB
28{
29 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
30 u64 target = 0;
31
32 if (!bctl)
33 return 0;
34
35 if (flags & BTRFS_BLOCK_GROUP_DATA &&
36 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
37 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
38 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
39 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
40 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
41 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
42 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
43 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
44 }
45
46 return target;
47}
48
49/*
50 * @flags: available profiles in extended format (see ctree.h)
51 *
52 * Return reduced profile in chunk format. If profile changing is in progress
53 * (either running or paused) picks the target profile (if it's already
54 * available), otherwise falls back to plain reducing.
55 */
56static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
57{
58 u64 num_devices = fs_info->fs_devices->rw_devices;
59 u64 target;
60 u64 raid_type;
61 u64 allowed = 0;
62
63 /*
64 * See if restripe for this chunk_type is in progress, if so try to
65 * reduce to the target profile
66 */
67 spin_lock(&fs_info->balance_lock);
e11c0406 68 target = get_restripe_target(fs_info, flags);
878d7b67 69 if (target) {
162e0a16
JB
70 spin_unlock(&fs_info->balance_lock);
71 return extended_to_chunk(target);
878d7b67
JB
72 }
73 spin_unlock(&fs_info->balance_lock);
74
75 /* First, mask out the RAID levels which aren't possible */
76 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
77 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
78 allowed |= btrfs_raid_array[raid_type].bg_flag;
79 }
80 allowed &= flags;
81
82 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
83 allowed = BTRFS_BLOCK_GROUP_RAID6;
84 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
85 allowed = BTRFS_BLOCK_GROUP_RAID5;
86 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
87 allowed = BTRFS_BLOCK_GROUP_RAID10;
88 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
89 allowed = BTRFS_BLOCK_GROUP_RAID1;
90 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
91 allowed = BTRFS_BLOCK_GROUP_RAID0;
92
93 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
94
95 return extended_to_chunk(flags | allowed);
96}
97
ef0a82da 98u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
878d7b67
JB
99{
100 unsigned seq;
101 u64 flags;
102
103 do {
104 flags = orig_flags;
105 seq = read_seqbegin(&fs_info->profiles_lock);
106
107 if (flags & BTRFS_BLOCK_GROUP_DATA)
108 flags |= fs_info->avail_data_alloc_bits;
109 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
110 flags |= fs_info->avail_system_alloc_bits;
111 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
112 flags |= fs_info->avail_metadata_alloc_bits;
113 } while (read_seqretry(&fs_info->profiles_lock, seq));
114
115 return btrfs_reduce_alloc_profile(fs_info, flags);
116}
117
32da5386 118void btrfs_get_block_group(struct btrfs_block_group *cache)
3cad1284 119{
48aaeebe 120 refcount_inc(&cache->refs);
3cad1284
JB
121}
122
32da5386 123void btrfs_put_block_group(struct btrfs_block_group *cache)
3cad1284 124{
48aaeebe 125 if (refcount_dec_and_test(&cache->refs)) {
3cad1284 126 WARN_ON(cache->pinned > 0);
40cdc509
FM
127 /*
128 * If there was a failure to cleanup a log tree, very likely due
129 * to an IO failure on a writeback attempt of one or more of its
130 * extent buffers, we could not do proper (and cheap) unaccounting
131 * of their reserved space, so don't warn on reserved > 0 in that
132 * case.
133 */
134 if (!(cache->flags & BTRFS_BLOCK_GROUP_METADATA) ||
135 !BTRFS_FS_LOG_CLEANUP_ERROR(cache->fs_info))
136 WARN_ON(cache->reserved > 0);
3cad1284 137
b0643e59
DZ
138 /*
139 * A block_group shouldn't be on the discard_list anymore.
140 * Remove the block_group from the discard_list to prevent us
141 * from causing a panic due to NULL pointer dereference.
142 */
143 if (WARN_ON(!list_empty(&cache->discard_list)))
144 btrfs_discard_cancel_work(&cache->fs_info->discard_ctl,
145 cache);
146
3cad1284
JB
147 /*
148 * If not empty, someone is still holding mutex of
149 * full_stripe_lock, which can only be released by caller.
150 * And it will definitely cause use-after-free when caller
151 * tries to release full stripe lock.
152 *
153 * No better way to resolve, but only to warn.
154 */
155 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
156 kfree(cache->free_space_ctl);
dafc340d 157 kfree(cache->physical_map);
3cad1284
JB
158 kfree(cache);
159 }
160}
161
4358d963
JB
162/*
163 * This adds the block group to the fs_info rb tree for the block group cache
164 */
165static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
32da5386 166 struct btrfs_block_group *block_group)
4358d963
JB
167{
168 struct rb_node **p;
169 struct rb_node *parent = NULL;
32da5386 170 struct btrfs_block_group *cache;
08dddb29 171 bool leftmost = true;
4358d963 172
9afc6649
QW
173 ASSERT(block_group->length != 0);
174
16b0c258 175 write_lock(&info->block_group_cache_lock);
08dddb29 176 p = &info->block_group_cache_tree.rb_root.rb_node;
4358d963
JB
177
178 while (*p) {
179 parent = *p;
32da5386 180 cache = rb_entry(parent, struct btrfs_block_group, cache_node);
b3470b5d 181 if (block_group->start < cache->start) {
4358d963 182 p = &(*p)->rb_left;
b3470b5d 183 } else if (block_group->start > cache->start) {
4358d963 184 p = &(*p)->rb_right;
08dddb29 185 leftmost = false;
4358d963 186 } else {
16b0c258 187 write_unlock(&info->block_group_cache_lock);
4358d963
JB
188 return -EEXIST;
189 }
190 }
191
192 rb_link_node(&block_group->cache_node, parent, p);
08dddb29
FM
193 rb_insert_color_cached(&block_group->cache_node,
194 &info->block_group_cache_tree, leftmost);
4358d963 195
16b0c258 196 write_unlock(&info->block_group_cache_lock);
4358d963
JB
197
198 return 0;
199}
200
2e405ad8
JB
201/*
202 * This will return the block group at or after bytenr if contains is 0, else
203 * it will return the block group that contains the bytenr
204 */
32da5386 205static struct btrfs_block_group *block_group_cache_tree_search(
2e405ad8
JB
206 struct btrfs_fs_info *info, u64 bytenr, int contains)
207{
32da5386 208 struct btrfs_block_group *cache, *ret = NULL;
2e405ad8
JB
209 struct rb_node *n;
210 u64 end, start;
211
16b0c258 212 read_lock(&info->block_group_cache_lock);
08dddb29 213 n = info->block_group_cache_tree.rb_root.rb_node;
2e405ad8
JB
214
215 while (n) {
32da5386 216 cache = rb_entry(n, struct btrfs_block_group, cache_node);
b3470b5d
DS
217 end = cache->start + cache->length - 1;
218 start = cache->start;
2e405ad8
JB
219
220 if (bytenr < start) {
b3470b5d 221 if (!contains && (!ret || start < ret->start))
2e405ad8
JB
222 ret = cache;
223 n = n->rb_left;
224 } else if (bytenr > start) {
225 if (contains && bytenr <= end) {
226 ret = cache;
227 break;
228 }
229 n = n->rb_right;
230 } else {
231 ret = cache;
232 break;
233 }
234 }
08dddb29 235 if (ret)
2e405ad8 236 btrfs_get_block_group(ret);
16b0c258 237 read_unlock(&info->block_group_cache_lock);
2e405ad8
JB
238
239 return ret;
240}
241
242/*
243 * Return the block group that starts at or after bytenr
244 */
32da5386 245struct btrfs_block_group *btrfs_lookup_first_block_group(
2e405ad8
JB
246 struct btrfs_fs_info *info, u64 bytenr)
247{
248 return block_group_cache_tree_search(info, bytenr, 0);
249}
250
251/*
252 * Return the block group that contains the given bytenr
253 */
32da5386 254struct btrfs_block_group *btrfs_lookup_block_group(
2e405ad8
JB
255 struct btrfs_fs_info *info, u64 bytenr)
256{
257 return block_group_cache_tree_search(info, bytenr, 1);
258}
259
32da5386
DS
260struct btrfs_block_group *btrfs_next_block_group(
261 struct btrfs_block_group *cache)
2e405ad8
JB
262{
263 struct btrfs_fs_info *fs_info = cache->fs_info;
264 struct rb_node *node;
265
16b0c258 266 read_lock(&fs_info->block_group_cache_lock);
2e405ad8
JB
267
268 /* If our block group was removed, we need a full search. */
269 if (RB_EMPTY_NODE(&cache->cache_node)) {
b3470b5d 270 const u64 next_bytenr = cache->start + cache->length;
2e405ad8 271
16b0c258 272 read_unlock(&fs_info->block_group_cache_lock);
2e405ad8 273 btrfs_put_block_group(cache);
8b01f931 274 return btrfs_lookup_first_block_group(fs_info, next_bytenr);
2e405ad8
JB
275 }
276 node = rb_next(&cache->cache_node);
277 btrfs_put_block_group(cache);
278 if (node) {
32da5386 279 cache = rb_entry(node, struct btrfs_block_group, cache_node);
2e405ad8
JB
280 btrfs_get_block_group(cache);
281 } else
282 cache = NULL;
16b0c258 283 read_unlock(&fs_info->block_group_cache_lock);
2e405ad8
JB
284 return cache;
285}
3eeb3226 286
2306e83e
FM
287/**
288 * Check if we can do a NOCOW write for a given extent.
289 *
290 * @fs_info: The filesystem information object.
291 * @bytenr: Logical start address of the extent.
292 *
293 * Check if we can do a NOCOW write for the given extent, and increments the
294 * number of NOCOW writers in the block group that contains the extent, as long
295 * as the block group exists and it's currently not in read-only mode.
296 *
297 * Returns: A non-NULL block group pointer if we can do a NOCOW write, the caller
298 * is responsible for calling btrfs_dec_nocow_writers() later.
299 *
300 * Or NULL if we can not do a NOCOW write
301 */
302struct btrfs_block_group *btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info,
303 u64 bytenr)
3eeb3226 304{
32da5386 305 struct btrfs_block_group *bg;
2306e83e 306 bool can_nocow = true;
3eeb3226
JB
307
308 bg = btrfs_lookup_block_group(fs_info, bytenr);
309 if (!bg)
2306e83e 310 return NULL;
3eeb3226
JB
311
312 spin_lock(&bg->lock);
313 if (bg->ro)
2306e83e 314 can_nocow = false;
3eeb3226
JB
315 else
316 atomic_inc(&bg->nocow_writers);
317 spin_unlock(&bg->lock);
318
2306e83e 319 if (!can_nocow) {
3eeb3226 320 btrfs_put_block_group(bg);
2306e83e
FM
321 return NULL;
322 }
3eeb3226 323
2306e83e
FM
324 /* No put on block group, done by btrfs_dec_nocow_writers(). */
325 return bg;
3eeb3226
JB
326}
327
2306e83e
FM
328/**
329 * Decrement the number of NOCOW writers in a block group.
330 *
331 * @bg: The block group.
332 *
333 * This is meant to be called after a previous call to btrfs_inc_nocow_writers(),
334 * and on the block group returned by that call. Typically this is called after
335 * creating an ordered extent for a NOCOW write, to prevent races with scrub and
336 * relocation.
337 *
338 * After this call, the caller should not use the block group anymore. It it wants
339 * to use it, then it should get a reference on it before calling this function.
340 */
341void btrfs_dec_nocow_writers(struct btrfs_block_group *bg)
3eeb3226 342{
3eeb3226
JB
343 if (atomic_dec_and_test(&bg->nocow_writers))
344 wake_up_var(&bg->nocow_writers);
2306e83e
FM
345
346 /* For the lookup done by a previous call to btrfs_inc_nocow_writers(). */
3eeb3226
JB
347 btrfs_put_block_group(bg);
348}
349
32da5386 350void btrfs_wait_nocow_writers(struct btrfs_block_group *bg)
3eeb3226
JB
351{
352 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
353}
354
355void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
356 const u64 start)
357{
32da5386 358 struct btrfs_block_group *bg;
3eeb3226
JB
359
360 bg = btrfs_lookup_block_group(fs_info, start);
361 ASSERT(bg);
362 if (atomic_dec_and_test(&bg->reservations))
363 wake_up_var(&bg->reservations);
364 btrfs_put_block_group(bg);
365}
366
32da5386 367void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg)
3eeb3226
JB
368{
369 struct btrfs_space_info *space_info = bg->space_info;
370
371 ASSERT(bg->ro);
372
373 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
374 return;
375
376 /*
377 * Our block group is read only but before we set it to read only,
378 * some task might have had allocated an extent from it already, but it
379 * has not yet created a respective ordered extent (and added it to a
380 * root's list of ordered extents).
381 * Therefore wait for any task currently allocating extents, since the
382 * block group's reservations counter is incremented while a read lock
383 * on the groups' semaphore is held and decremented after releasing
384 * the read access on that semaphore and creating the ordered extent.
385 */
386 down_write(&space_info->groups_sem);
387 up_write(&space_info->groups_sem);
388
389 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
390}
9f21246d
JB
391
392struct btrfs_caching_control *btrfs_get_caching_control(
32da5386 393 struct btrfs_block_group *cache)
9f21246d
JB
394{
395 struct btrfs_caching_control *ctl;
396
397 spin_lock(&cache->lock);
398 if (!cache->caching_ctl) {
399 spin_unlock(&cache->lock);
400 return NULL;
401 }
402
403 ctl = cache->caching_ctl;
404 refcount_inc(&ctl->count);
405 spin_unlock(&cache->lock);
406 return ctl;
407}
408
409void btrfs_put_caching_control(struct btrfs_caching_control *ctl)
410{
411 if (refcount_dec_and_test(&ctl->count))
412 kfree(ctl);
413}
414
415/*
416 * When we wait for progress in the block group caching, its because our
417 * allocation attempt failed at least once. So, we must sleep and let some
418 * progress happen before we try again.
419 *
420 * This function will sleep at least once waiting for new free space to show
421 * up, and then it will check the block group free space numbers for our min
422 * num_bytes. Another option is to have it go ahead and look in the rbtree for
423 * a free extent of a given size, but this is a good start.
424 *
425 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
426 * any of the information in this block group.
427 */
32da5386 428void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache,
9f21246d
JB
429 u64 num_bytes)
430{
431 struct btrfs_caching_control *caching_ctl;
432
433 caching_ctl = btrfs_get_caching_control(cache);
434 if (!caching_ctl)
435 return;
436
32da5386 437 wait_event(caching_ctl->wait, btrfs_block_group_done(cache) ||
9f21246d
JB
438 (cache->free_space_ctl->free_space >= num_bytes));
439
440 btrfs_put_caching_control(caching_ctl);
441}
442
ced8ecf0
OS
443static int btrfs_caching_ctl_wait_done(struct btrfs_block_group *cache,
444 struct btrfs_caching_control *caching_ctl)
445{
446 wait_event(caching_ctl->wait, btrfs_block_group_done(cache));
447 return cache->cached == BTRFS_CACHE_ERROR ? -EIO : 0;
448}
449
450static int btrfs_wait_block_group_cache_done(struct btrfs_block_group *cache)
9f21246d
JB
451{
452 struct btrfs_caching_control *caching_ctl;
ced8ecf0 453 int ret;
9f21246d
JB
454
455 caching_ctl = btrfs_get_caching_control(cache);
456 if (!caching_ctl)
457 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
ced8ecf0 458 ret = btrfs_caching_ctl_wait_done(cache, caching_ctl);
9f21246d
JB
459 btrfs_put_caching_control(caching_ctl);
460 return ret;
461}
462
463#ifdef CONFIG_BTRFS_DEBUG
32da5386 464static void fragment_free_space(struct btrfs_block_group *block_group)
9f21246d
JB
465{
466 struct btrfs_fs_info *fs_info = block_group->fs_info;
b3470b5d
DS
467 u64 start = block_group->start;
468 u64 len = block_group->length;
9f21246d
JB
469 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
470 fs_info->nodesize : fs_info->sectorsize;
471 u64 step = chunk << 1;
472
473 while (len > chunk) {
474 btrfs_remove_free_space(block_group, start, chunk);
475 start += step;
476 if (len < step)
477 len = 0;
478 else
479 len -= step;
480 }
481}
482#endif
483
484/*
485 * This is only called by btrfs_cache_block_group, since we could have freed
486 * extents we need to check the pinned_extents for any extents that can't be
487 * used yet since their free space will be released as soon as the transaction
488 * commits.
489 */
32da5386 490u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end)
9f21246d
JB
491{
492 struct btrfs_fs_info *info = block_group->fs_info;
493 u64 extent_start, extent_end, size, total_added = 0;
494 int ret;
495
496 while (start < end) {
fe119a6e 497 ret = find_first_extent_bit(&info->excluded_extents, start,
9f21246d
JB
498 &extent_start, &extent_end,
499 EXTENT_DIRTY | EXTENT_UPTODATE,
500 NULL);
501 if (ret)
502 break;
503
504 if (extent_start <= start) {
505 start = extent_end + 1;
506 } else if (extent_start > start && extent_start < end) {
507 size = extent_start - start;
508 total_added += size;
b0643e59
DZ
509 ret = btrfs_add_free_space_async_trimmed(block_group,
510 start, size);
9f21246d
JB
511 BUG_ON(ret); /* -ENOMEM or logic error */
512 start = extent_end + 1;
513 } else {
514 break;
515 }
516 }
517
518 if (start < end) {
519 size = end - start;
520 total_added += size;
b0643e59
DZ
521 ret = btrfs_add_free_space_async_trimmed(block_group, start,
522 size);
9f21246d
JB
523 BUG_ON(ret); /* -ENOMEM or logic error */
524 }
525
526 return total_added;
527}
528
529static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
530{
32da5386 531 struct btrfs_block_group *block_group = caching_ctl->block_group;
9f21246d 532 struct btrfs_fs_info *fs_info = block_group->fs_info;
29cbcf40 533 struct btrfs_root *extent_root;
9f21246d
JB
534 struct btrfs_path *path;
535 struct extent_buffer *leaf;
536 struct btrfs_key key;
537 u64 total_found = 0;
538 u64 last = 0;
539 u32 nritems;
540 int ret;
541 bool wakeup = true;
542
543 path = btrfs_alloc_path();
544 if (!path)
545 return -ENOMEM;
546
b3470b5d 547 last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET);
29cbcf40 548 extent_root = btrfs_extent_root(fs_info, last);
9f21246d
JB
549
550#ifdef CONFIG_BTRFS_DEBUG
551 /*
552 * If we're fragmenting we don't want to make anybody think we can
553 * allocate from this block group until we've had a chance to fragment
554 * the free space.
555 */
556 if (btrfs_should_fragment_free_space(block_group))
557 wakeup = false;
558#endif
559 /*
560 * We don't want to deadlock with somebody trying to allocate a new
561 * extent for the extent root while also trying to search the extent
562 * root to add free space. So we skip locking and search the commit
563 * root, since its read-only
564 */
565 path->skip_locking = 1;
566 path->search_commit_root = 1;
567 path->reada = READA_FORWARD;
568
569 key.objectid = last;
570 key.offset = 0;
571 key.type = BTRFS_EXTENT_ITEM_KEY;
572
573next:
574 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
575 if (ret < 0)
576 goto out;
577
578 leaf = path->nodes[0];
579 nritems = btrfs_header_nritems(leaf);
580
581 while (1) {
582 if (btrfs_fs_closing(fs_info) > 1) {
583 last = (u64)-1;
584 break;
585 }
586
587 if (path->slots[0] < nritems) {
588 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
589 } else {
590 ret = btrfs_find_next_key(extent_root, path, &key, 0, 0);
591 if (ret)
592 break;
593
594 if (need_resched() ||
595 rwsem_is_contended(&fs_info->commit_root_sem)) {
596 if (wakeup)
597 caching_ctl->progress = last;
598 btrfs_release_path(path);
599 up_read(&fs_info->commit_root_sem);
600 mutex_unlock(&caching_ctl->mutex);
601 cond_resched();
602 mutex_lock(&caching_ctl->mutex);
603 down_read(&fs_info->commit_root_sem);
604 goto next;
605 }
606
607 ret = btrfs_next_leaf(extent_root, path);
608 if (ret < 0)
609 goto out;
610 if (ret)
611 break;
612 leaf = path->nodes[0];
613 nritems = btrfs_header_nritems(leaf);
614 continue;
615 }
616
617 if (key.objectid < last) {
618 key.objectid = last;
619 key.offset = 0;
620 key.type = BTRFS_EXTENT_ITEM_KEY;
621
622 if (wakeup)
623 caching_ctl->progress = last;
624 btrfs_release_path(path);
625 goto next;
626 }
627
b3470b5d 628 if (key.objectid < block_group->start) {
9f21246d
JB
629 path->slots[0]++;
630 continue;
631 }
632
b3470b5d 633 if (key.objectid >= block_group->start + block_group->length)
9f21246d
JB
634 break;
635
636 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
637 key.type == BTRFS_METADATA_ITEM_KEY) {
638 total_found += add_new_free_space(block_group, last,
639 key.objectid);
640 if (key.type == BTRFS_METADATA_ITEM_KEY)
641 last = key.objectid +
642 fs_info->nodesize;
643 else
644 last = key.objectid + key.offset;
645
646 if (total_found > CACHING_CTL_WAKE_UP) {
647 total_found = 0;
648 if (wakeup)
649 wake_up(&caching_ctl->wait);
650 }
651 }
652 path->slots[0]++;
653 }
654 ret = 0;
655
656 total_found += add_new_free_space(block_group, last,
b3470b5d 657 block_group->start + block_group->length);
9f21246d
JB
658 caching_ctl->progress = (u64)-1;
659
660out:
661 btrfs_free_path(path);
662 return ret;
663}
664
665static noinline void caching_thread(struct btrfs_work *work)
666{
32da5386 667 struct btrfs_block_group *block_group;
9f21246d
JB
668 struct btrfs_fs_info *fs_info;
669 struct btrfs_caching_control *caching_ctl;
670 int ret;
671
672 caching_ctl = container_of(work, struct btrfs_caching_control, work);
673 block_group = caching_ctl->block_group;
674 fs_info = block_group->fs_info;
675
676 mutex_lock(&caching_ctl->mutex);
677 down_read(&fs_info->commit_root_sem);
678
e747853c
JB
679 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
680 ret = load_free_space_cache(block_group);
681 if (ret == 1) {
682 ret = 0;
683 goto done;
684 }
685
686 /*
687 * We failed to load the space cache, set ourselves to
688 * CACHE_STARTED and carry on.
689 */
690 spin_lock(&block_group->lock);
691 block_group->cached = BTRFS_CACHE_STARTED;
692 spin_unlock(&block_group->lock);
693 wake_up(&caching_ctl->wait);
694 }
695
2f96e402
JB
696 /*
697 * If we are in the transaction that populated the free space tree we
698 * can't actually cache from the free space tree as our commit root and
699 * real root are the same, so we could change the contents of the blocks
700 * while caching. Instead do the slow caching in this case, and after
701 * the transaction has committed we will be safe.
702 */
703 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
704 !(test_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags)))
9f21246d
JB
705 ret = load_free_space_tree(caching_ctl);
706 else
707 ret = load_extent_tree_free(caching_ctl);
e747853c 708done:
9f21246d
JB
709 spin_lock(&block_group->lock);
710 block_group->caching_ctl = NULL;
711 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
712 spin_unlock(&block_group->lock);
713
714#ifdef CONFIG_BTRFS_DEBUG
715 if (btrfs_should_fragment_free_space(block_group)) {
716 u64 bytes_used;
717
718 spin_lock(&block_group->space_info->lock);
719 spin_lock(&block_group->lock);
b3470b5d 720 bytes_used = block_group->length - block_group->used;
9f21246d
JB
721 block_group->space_info->bytes_used += bytes_used >> 1;
722 spin_unlock(&block_group->lock);
723 spin_unlock(&block_group->space_info->lock);
e11c0406 724 fragment_free_space(block_group);
9f21246d
JB
725 }
726#endif
727
728 caching_ctl->progress = (u64)-1;
729
730 up_read(&fs_info->commit_root_sem);
731 btrfs_free_excluded_extents(block_group);
732 mutex_unlock(&caching_ctl->mutex);
733
734 wake_up(&caching_ctl->wait);
735
736 btrfs_put_caching_control(caching_ctl);
737 btrfs_put_block_group(block_group);
738}
739
ced8ecf0 740int btrfs_cache_block_group(struct btrfs_block_group *cache, bool wait)
9f21246d 741{
9f21246d 742 struct btrfs_fs_info *fs_info = cache->fs_info;
e747853c 743 struct btrfs_caching_control *caching_ctl = NULL;
9f21246d
JB
744 int ret = 0;
745
2eda5708
NA
746 /* Allocator for zoned filesystems does not use the cache at all */
747 if (btrfs_is_zoned(fs_info))
748 return 0;
749
9f21246d
JB
750 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
751 if (!caching_ctl)
752 return -ENOMEM;
753
754 INIT_LIST_HEAD(&caching_ctl->list);
755 mutex_init(&caching_ctl->mutex);
756 init_waitqueue_head(&caching_ctl->wait);
757 caching_ctl->block_group = cache;
b3470b5d 758 caching_ctl->progress = cache->start;
e747853c 759 refcount_set(&caching_ctl->count, 2);
a0cac0ec 760 btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL);
9f21246d
JB
761
762 spin_lock(&cache->lock);
9f21246d 763 if (cache->cached != BTRFS_CACHE_NO) {
9f21246d 764 kfree(caching_ctl);
e747853c
JB
765
766 caching_ctl = cache->caching_ctl;
767 if (caching_ctl)
768 refcount_inc(&caching_ctl->count);
769 spin_unlock(&cache->lock);
770 goto out;
9f21246d
JB
771 }
772 WARN_ON(cache->caching_ctl);
773 cache->caching_ctl = caching_ctl;
ced8ecf0 774 cache->cached = BTRFS_CACHE_STARTED;
e747853c 775 cache->has_caching_ctl = 1;
9f21246d
JB
776 spin_unlock(&cache->lock);
777
16b0c258 778 write_lock(&fs_info->block_group_cache_lock);
9f21246d
JB
779 refcount_inc(&caching_ctl->count);
780 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
16b0c258 781 write_unlock(&fs_info->block_group_cache_lock);
9f21246d
JB
782
783 btrfs_get_block_group(cache);
784
785 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
e747853c 786out:
ced8ecf0
OS
787 if (wait && caching_ctl)
788 ret = btrfs_caching_ctl_wait_done(cache, caching_ctl);
e747853c
JB
789 if (caching_ctl)
790 btrfs_put_caching_control(caching_ctl);
9f21246d
JB
791
792 return ret;
793}
e3e0520b
JB
794
795static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
796{
797 u64 extra_flags = chunk_to_extended(flags) &
798 BTRFS_EXTENDED_PROFILE_MASK;
799
800 write_seqlock(&fs_info->profiles_lock);
801 if (flags & BTRFS_BLOCK_GROUP_DATA)
802 fs_info->avail_data_alloc_bits &= ~extra_flags;
803 if (flags & BTRFS_BLOCK_GROUP_METADATA)
804 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
805 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
806 fs_info->avail_system_alloc_bits &= ~extra_flags;
807 write_sequnlock(&fs_info->profiles_lock);
808}
809
810/*
811 * Clear incompat bits for the following feature(s):
812 *
813 * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group
814 * in the whole filesystem
9c907446
DS
815 *
816 * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups
e3e0520b
JB
817 */
818static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags)
819{
9c907446
DS
820 bool found_raid56 = false;
821 bool found_raid1c34 = false;
822
823 if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) ||
824 (flags & BTRFS_BLOCK_GROUP_RAID1C3) ||
825 (flags & BTRFS_BLOCK_GROUP_RAID1C4)) {
e3e0520b
JB
826 struct list_head *head = &fs_info->space_info;
827 struct btrfs_space_info *sinfo;
828
829 list_for_each_entry_rcu(sinfo, head, list) {
e3e0520b
JB
830 down_read(&sinfo->groups_sem);
831 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5]))
9c907446 832 found_raid56 = true;
e3e0520b 833 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6]))
9c907446
DS
834 found_raid56 = true;
835 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3]))
836 found_raid1c34 = true;
837 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4]))
838 found_raid1c34 = true;
e3e0520b 839 up_read(&sinfo->groups_sem);
e3e0520b 840 }
d8e6fd5c 841 if (!found_raid56)
9c907446 842 btrfs_clear_fs_incompat(fs_info, RAID56);
d8e6fd5c 843 if (!found_raid1c34)
9c907446 844 btrfs_clear_fs_incompat(fs_info, RAID1C34);
e3e0520b
JB
845 }
846}
847
7357623a
QW
848static int remove_block_group_item(struct btrfs_trans_handle *trans,
849 struct btrfs_path *path,
850 struct btrfs_block_group *block_group)
851{
852 struct btrfs_fs_info *fs_info = trans->fs_info;
853 struct btrfs_root *root;
854 struct btrfs_key key;
855 int ret;
856
dfe8aec4 857 root = btrfs_block_group_root(fs_info);
7357623a
QW
858 key.objectid = block_group->start;
859 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
860 key.offset = block_group->length;
861
862 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
863 if (ret > 0)
864 ret = -ENOENT;
865 if (ret < 0)
866 return ret;
867
868 ret = btrfs_del_item(trans, root, path);
869 return ret;
870}
871
e3e0520b
JB
872int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
873 u64 group_start, struct extent_map *em)
874{
875 struct btrfs_fs_info *fs_info = trans->fs_info;
e3e0520b 876 struct btrfs_path *path;
32da5386 877 struct btrfs_block_group *block_group;
e3e0520b 878 struct btrfs_free_cluster *cluster;
e3e0520b
JB
879 struct inode *inode;
880 struct kobject *kobj = NULL;
881 int ret;
882 int index;
883 int factor;
884 struct btrfs_caching_control *caching_ctl = NULL;
885 bool remove_em;
886 bool remove_rsv = false;
887
888 block_group = btrfs_lookup_block_group(fs_info, group_start);
889 BUG_ON(!block_group);
890 BUG_ON(!block_group->ro);
891
892 trace_btrfs_remove_block_group(block_group);
893 /*
894 * Free the reserved super bytes from this block group before
895 * remove it.
896 */
897 btrfs_free_excluded_extents(block_group);
b3470b5d
DS
898 btrfs_free_ref_tree_range(fs_info, block_group->start,
899 block_group->length);
e3e0520b 900
e3e0520b
JB
901 index = btrfs_bg_flags_to_raid_index(block_group->flags);
902 factor = btrfs_bg_type_to_factor(block_group->flags);
903
904 /* make sure this block group isn't part of an allocation cluster */
905 cluster = &fs_info->data_alloc_cluster;
906 spin_lock(&cluster->refill_lock);
907 btrfs_return_cluster_to_free_space(block_group, cluster);
908 spin_unlock(&cluster->refill_lock);
909
910 /*
911 * make sure this block group isn't part of a metadata
912 * allocation cluster
913 */
914 cluster = &fs_info->meta_alloc_cluster;
915 spin_lock(&cluster->refill_lock);
916 btrfs_return_cluster_to_free_space(block_group, cluster);
917 spin_unlock(&cluster->refill_lock);
918
40ab3be1 919 btrfs_clear_treelog_bg(block_group);
c2707a25 920 btrfs_clear_data_reloc_bg(block_group);
40ab3be1 921
e3e0520b
JB
922 path = btrfs_alloc_path();
923 if (!path) {
924 ret = -ENOMEM;
9fecd132 925 goto out;
e3e0520b
JB
926 }
927
928 /*
929 * get the inode first so any iput calls done for the io_list
930 * aren't the final iput (no unlinks allowed now)
931 */
932 inode = lookup_free_space_inode(block_group, path);
933
934 mutex_lock(&trans->transaction->cache_write_mutex);
935 /*
936 * Make sure our free space cache IO is done before removing the
937 * free space inode
938 */
939 spin_lock(&trans->transaction->dirty_bgs_lock);
940 if (!list_empty(&block_group->io_list)) {
941 list_del_init(&block_group->io_list);
942
943 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
944
945 spin_unlock(&trans->transaction->dirty_bgs_lock);
946 btrfs_wait_cache_io(trans, block_group, path);
947 btrfs_put_block_group(block_group);
948 spin_lock(&trans->transaction->dirty_bgs_lock);
949 }
950
951 if (!list_empty(&block_group->dirty_list)) {
952 list_del_init(&block_group->dirty_list);
953 remove_rsv = true;
954 btrfs_put_block_group(block_group);
955 }
956 spin_unlock(&trans->transaction->dirty_bgs_lock);
957 mutex_unlock(&trans->transaction->cache_write_mutex);
958
36b216c8
BB
959 ret = btrfs_remove_free_space_inode(trans, inode, block_group);
960 if (ret)
9fecd132 961 goto out;
e3e0520b 962
16b0c258 963 write_lock(&fs_info->block_group_cache_lock);
08dddb29
FM
964 rb_erase_cached(&block_group->cache_node,
965 &fs_info->block_group_cache_tree);
e3e0520b
JB
966 RB_CLEAR_NODE(&block_group->cache_node);
967
9fecd132
FM
968 /* Once for the block groups rbtree */
969 btrfs_put_block_group(block_group);
970
16b0c258 971 write_unlock(&fs_info->block_group_cache_lock);
e3e0520b
JB
972
973 down_write(&block_group->space_info->groups_sem);
974 /*
975 * we must use list_del_init so people can check to see if they
976 * are still on the list after taking the semaphore
977 */
978 list_del_init(&block_group->list);
979 if (list_empty(&block_group->space_info->block_groups[index])) {
980 kobj = block_group->space_info->block_group_kobjs[index];
981 block_group->space_info->block_group_kobjs[index] = NULL;
982 clear_avail_alloc_bits(fs_info, block_group->flags);
983 }
984 up_write(&block_group->space_info->groups_sem);
985 clear_incompat_bg_bits(fs_info, block_group->flags);
986 if (kobj) {
987 kobject_del(kobj);
988 kobject_put(kobj);
989 }
990
991 if (block_group->has_caching_ctl)
992 caching_ctl = btrfs_get_caching_control(block_group);
993 if (block_group->cached == BTRFS_CACHE_STARTED)
994 btrfs_wait_block_group_cache_done(block_group);
995 if (block_group->has_caching_ctl) {
16b0c258 996 write_lock(&fs_info->block_group_cache_lock);
e3e0520b
JB
997 if (!caching_ctl) {
998 struct btrfs_caching_control *ctl;
999
1000 list_for_each_entry(ctl,
1001 &fs_info->caching_block_groups, list)
1002 if (ctl->block_group == block_group) {
1003 caching_ctl = ctl;
1004 refcount_inc(&caching_ctl->count);
1005 break;
1006 }
1007 }
1008 if (caching_ctl)
1009 list_del_init(&caching_ctl->list);
16b0c258 1010 write_unlock(&fs_info->block_group_cache_lock);
e3e0520b
JB
1011 if (caching_ctl) {
1012 /* Once for the caching bgs list and once for us. */
1013 btrfs_put_caching_control(caching_ctl);
1014 btrfs_put_caching_control(caching_ctl);
1015 }
1016 }
1017
1018 spin_lock(&trans->transaction->dirty_bgs_lock);
1019 WARN_ON(!list_empty(&block_group->dirty_list));
1020 WARN_ON(!list_empty(&block_group->io_list));
1021 spin_unlock(&trans->transaction->dirty_bgs_lock);
1022
1023 btrfs_remove_free_space_cache(block_group);
1024
1025 spin_lock(&block_group->space_info->lock);
1026 list_del_init(&block_group->ro_list);
1027
1028 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
1029 WARN_ON(block_group->space_info->total_bytes
b3470b5d 1030 < block_group->length);
e3e0520b 1031 WARN_ON(block_group->space_info->bytes_readonly
169e0da9
NA
1032 < block_group->length - block_group->zone_unusable);
1033 WARN_ON(block_group->space_info->bytes_zone_unusable
1034 < block_group->zone_unusable);
e3e0520b 1035 WARN_ON(block_group->space_info->disk_total
b3470b5d 1036 < block_group->length * factor);
6a921de5
NA
1037 WARN_ON(block_group->zone_is_active &&
1038 block_group->space_info->active_total_bytes
1039 < block_group->length);
e3e0520b 1040 }
b3470b5d 1041 block_group->space_info->total_bytes -= block_group->length;
6a921de5
NA
1042 if (block_group->zone_is_active)
1043 block_group->space_info->active_total_bytes -= block_group->length;
169e0da9
NA
1044 block_group->space_info->bytes_readonly -=
1045 (block_group->length - block_group->zone_unusable);
1046 block_group->space_info->bytes_zone_unusable -=
1047 block_group->zone_unusable;
b3470b5d 1048 block_group->space_info->disk_total -= block_group->length * factor;
e3e0520b
JB
1049
1050 spin_unlock(&block_group->space_info->lock);
1051
ffcb9d44
FM
1052 /*
1053 * Remove the free space for the block group from the free space tree
1054 * and the block group's item from the extent tree before marking the
1055 * block group as removed. This is to prevent races with tasks that
1056 * freeze and unfreeze a block group, this task and another task
1057 * allocating a new block group - the unfreeze task ends up removing
1058 * the block group's extent map before the task calling this function
1059 * deletes the block group item from the extent tree, allowing for
1060 * another task to attempt to create another block group with the same
1061 * item key (and failing with -EEXIST and a transaction abort).
1062 */
1063 ret = remove_block_group_free_space(trans, block_group);
1064 if (ret)
1065 goto out;
1066
1067 ret = remove_block_group_item(trans, path, block_group);
1068 if (ret < 0)
1069 goto out;
1070
e3e0520b
JB
1071 spin_lock(&block_group->lock);
1072 block_group->removed = 1;
1073 /*
6b7304af
FM
1074 * At this point trimming or scrub can't start on this block group,
1075 * because we removed the block group from the rbtree
1076 * fs_info->block_group_cache_tree so no one can't find it anymore and
1077 * even if someone already got this block group before we removed it
1078 * from the rbtree, they have already incremented block_group->frozen -
1079 * if they didn't, for the trimming case they won't find any free space
1080 * entries because we already removed them all when we called
1081 * btrfs_remove_free_space_cache().
e3e0520b
JB
1082 *
1083 * And we must not remove the extent map from the fs_info->mapping_tree
1084 * to prevent the same logical address range and physical device space
6b7304af
FM
1085 * ranges from being reused for a new block group. This is needed to
1086 * avoid races with trimming and scrub.
1087 *
1088 * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
e3e0520b
JB
1089 * completely transactionless, so while it is trimming a range the
1090 * currently running transaction might finish and a new one start,
1091 * allowing for new block groups to be created that can reuse the same
1092 * physical device locations unless we take this special care.
1093 *
1094 * There may also be an implicit trim operation if the file system
1095 * is mounted with -odiscard. The same protections must remain
1096 * in place until the extents have been discarded completely when
1097 * the transaction commit has completed.
1098 */
6b7304af 1099 remove_em = (atomic_read(&block_group->frozen) == 0);
e3e0520b
JB
1100 spin_unlock(&block_group->lock);
1101
e3e0520b
JB
1102 if (remove_em) {
1103 struct extent_map_tree *em_tree;
1104
1105 em_tree = &fs_info->mapping_tree;
1106 write_lock(&em_tree->lock);
1107 remove_extent_mapping(em_tree, em);
1108 write_unlock(&em_tree->lock);
1109 /* once for the tree */
1110 free_extent_map(em);
1111 }
f6033c5e 1112
9fecd132 1113out:
f6033c5e
XY
1114 /* Once for the lookup reference */
1115 btrfs_put_block_group(block_group);
e3e0520b
JB
1116 if (remove_rsv)
1117 btrfs_delayed_refs_rsv_release(fs_info, 1);
1118 btrfs_free_path(path);
1119 return ret;
1120}
1121
1122struct btrfs_trans_handle *btrfs_start_trans_remove_block_group(
1123 struct btrfs_fs_info *fs_info, const u64 chunk_offset)
1124{
dfe8aec4 1125 struct btrfs_root *root = btrfs_block_group_root(fs_info);
e3e0520b
JB
1126 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1127 struct extent_map *em;
1128 struct map_lookup *map;
1129 unsigned int num_items;
1130
1131 read_lock(&em_tree->lock);
1132 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1133 read_unlock(&em_tree->lock);
1134 ASSERT(em && em->start == chunk_offset);
1135
1136 /*
1137 * We need to reserve 3 + N units from the metadata space info in order
1138 * to remove a block group (done at btrfs_remove_chunk() and at
1139 * btrfs_remove_block_group()), which are used for:
1140 *
1141 * 1 unit for adding the free space inode's orphan (located in the tree
1142 * of tree roots).
1143 * 1 unit for deleting the block group item (located in the extent
1144 * tree).
1145 * 1 unit for deleting the free space item (located in tree of tree
1146 * roots).
1147 * N units for deleting N device extent items corresponding to each
1148 * stripe (located in the device tree).
1149 *
1150 * In order to remove a block group we also need to reserve units in the
1151 * system space info in order to update the chunk tree (update one or
1152 * more device items and remove one chunk item), but this is done at
1153 * btrfs_remove_chunk() through a call to check_system_chunk().
1154 */
1155 map = em->map_lookup;
1156 num_items = 3 + map->num_stripes;
1157 free_extent_map(em);
1158
dfe8aec4 1159 return btrfs_start_transaction_fallback_global_rsv(root, num_items);
e3e0520b
JB
1160}
1161
26ce2095
JB
1162/*
1163 * Mark block group @cache read-only, so later write won't happen to block
1164 * group @cache.
1165 *
1166 * If @force is not set, this function will only mark the block group readonly
1167 * if we have enough free space (1M) in other metadata/system block groups.
1168 * If @force is not set, this function will mark the block group readonly
1169 * without checking free space.
1170 *
1171 * NOTE: This function doesn't care if other block groups can contain all the
1172 * data in this block group. That check should be done by relocation routine,
1173 * not this function.
1174 */
32da5386 1175static int inc_block_group_ro(struct btrfs_block_group *cache, int force)
26ce2095
JB
1176{
1177 struct btrfs_space_info *sinfo = cache->space_info;
1178 u64 num_bytes;
26ce2095
JB
1179 int ret = -ENOSPC;
1180
26ce2095
JB
1181 spin_lock(&sinfo->lock);
1182 spin_lock(&cache->lock);
1183
195a49ea
FM
1184 if (cache->swap_extents) {
1185 ret = -ETXTBSY;
1186 goto out;
1187 }
1188
26ce2095
JB
1189 if (cache->ro) {
1190 cache->ro++;
1191 ret = 0;
1192 goto out;
1193 }
1194
b3470b5d 1195 num_bytes = cache->length - cache->reserved - cache->pinned -
169e0da9 1196 cache->bytes_super - cache->zone_unusable - cache->used;
26ce2095
JB
1197
1198 /*
a30a3d20
JB
1199 * Data never overcommits, even in mixed mode, so do just the straight
1200 * check of left over space in how much we have allocated.
26ce2095 1201 */
a30a3d20
JB
1202 if (force) {
1203 ret = 0;
1204 } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) {
1205 u64 sinfo_used = btrfs_space_info_used(sinfo, true);
1206
1207 /*
1208 * Here we make sure if we mark this bg RO, we still have enough
1209 * free space as buffer.
1210 */
1211 if (sinfo_used + num_bytes <= sinfo->total_bytes)
1212 ret = 0;
1213 } else {
1214 /*
1215 * We overcommit metadata, so we need to do the
1216 * btrfs_can_overcommit check here, and we need to pass in
1217 * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of
1218 * leeway to allow us to mark this block group as read only.
1219 */
1220 if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes,
1221 BTRFS_RESERVE_NO_FLUSH))
1222 ret = 0;
1223 }
1224
1225 if (!ret) {
26ce2095 1226 sinfo->bytes_readonly += num_bytes;
169e0da9
NA
1227 if (btrfs_is_zoned(cache->fs_info)) {
1228 /* Migrate zone_unusable bytes to readonly */
1229 sinfo->bytes_readonly += cache->zone_unusable;
1230 sinfo->bytes_zone_unusable -= cache->zone_unusable;
1231 cache->zone_unusable = 0;
1232 }
26ce2095
JB
1233 cache->ro++;
1234 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
26ce2095
JB
1235 }
1236out:
1237 spin_unlock(&cache->lock);
1238 spin_unlock(&sinfo->lock);
1239 if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) {
1240 btrfs_info(cache->fs_info,
b3470b5d 1241 "unable to make block group %llu ro", cache->start);
26ce2095
JB
1242 btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0);
1243 }
1244 return ret;
1245}
1246
fe119a6e
NB
1247static bool clean_pinned_extents(struct btrfs_trans_handle *trans,
1248 struct btrfs_block_group *bg)
45bb5d6a
NB
1249{
1250 struct btrfs_fs_info *fs_info = bg->fs_info;
fe119a6e 1251 struct btrfs_transaction *prev_trans = NULL;
45bb5d6a
NB
1252 const u64 start = bg->start;
1253 const u64 end = start + bg->length - 1;
1254 int ret;
1255
fe119a6e
NB
1256 spin_lock(&fs_info->trans_lock);
1257 if (trans->transaction->list.prev != &fs_info->trans_list) {
1258 prev_trans = list_last_entry(&trans->transaction->list,
1259 struct btrfs_transaction, list);
1260 refcount_inc(&prev_trans->use_count);
1261 }
1262 spin_unlock(&fs_info->trans_lock);
1263
45bb5d6a
NB
1264 /*
1265 * Hold the unused_bg_unpin_mutex lock to avoid racing with
1266 * btrfs_finish_extent_commit(). If we are at transaction N, another
1267 * task might be running finish_extent_commit() for the previous
1268 * transaction N - 1, and have seen a range belonging to the block
fe119a6e
NB
1269 * group in pinned_extents before we were able to clear the whole block
1270 * group range from pinned_extents. This means that task can lookup for
1271 * the block group after we unpinned it from pinned_extents and removed
1272 * it, leading to a BUG_ON() at unpin_extent_range().
45bb5d6a
NB
1273 */
1274 mutex_lock(&fs_info->unused_bg_unpin_mutex);
fe119a6e
NB
1275 if (prev_trans) {
1276 ret = clear_extent_bits(&prev_trans->pinned_extents, start, end,
1277 EXTENT_DIRTY);
1278 if (ret)
534cf531 1279 goto out;
fe119a6e 1280 }
45bb5d6a 1281
fe119a6e 1282 ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end,
45bb5d6a 1283 EXTENT_DIRTY);
534cf531 1284out:
45bb5d6a 1285 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5150bf19
FM
1286 if (prev_trans)
1287 btrfs_put_transaction(prev_trans);
45bb5d6a 1288
534cf531 1289 return ret == 0;
45bb5d6a
NB
1290}
1291
e3e0520b
JB
1292/*
1293 * Process the unused_bgs list and remove any that don't have any allocated
1294 * space inside of them.
1295 */
1296void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
1297{
32da5386 1298 struct btrfs_block_group *block_group;
e3e0520b
JB
1299 struct btrfs_space_info *space_info;
1300 struct btrfs_trans_handle *trans;
6e80d4f8 1301 const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC);
e3e0520b
JB
1302 int ret = 0;
1303
1304 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1305 return;
1306
2f12741f
JB
1307 if (btrfs_fs_closing(fs_info))
1308 return;
1309
ddfd08cb
JB
1310 /*
1311 * Long running balances can keep us blocked here for eternity, so
1312 * simply skip deletion if we're unable to get the mutex.
1313 */
f3372065 1314 if (!mutex_trylock(&fs_info->reclaim_bgs_lock))
ddfd08cb
JB
1315 return;
1316
e3e0520b
JB
1317 spin_lock(&fs_info->unused_bgs_lock);
1318 while (!list_empty(&fs_info->unused_bgs)) {
e3e0520b
JB
1319 int trimming;
1320
1321 block_group = list_first_entry(&fs_info->unused_bgs,
32da5386 1322 struct btrfs_block_group,
e3e0520b
JB
1323 bg_list);
1324 list_del_init(&block_group->bg_list);
1325
1326 space_info = block_group->space_info;
1327
1328 if (ret || btrfs_mixed_space_info(space_info)) {
1329 btrfs_put_block_group(block_group);
1330 continue;
1331 }
1332 spin_unlock(&fs_info->unused_bgs_lock);
1333
b0643e59
DZ
1334 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
1335
e3e0520b
JB
1336 /* Don't want to race with allocators so take the groups_sem */
1337 down_write(&space_info->groups_sem);
6e80d4f8
DZ
1338
1339 /*
1340 * Async discard moves the final block group discard to be prior
1341 * to the unused_bgs code path. Therefore, if it's not fully
1342 * trimmed, punt it back to the async discard lists.
1343 */
1344 if (btrfs_test_opt(fs_info, DISCARD_ASYNC) &&
1345 !btrfs_is_free_space_trimmed(block_group)) {
1346 trace_btrfs_skip_unused_block_group(block_group);
1347 up_write(&space_info->groups_sem);
1348 /* Requeue if we failed because of async discard */
1349 btrfs_discard_queue_work(&fs_info->discard_ctl,
1350 block_group);
1351 goto next;
1352 }
1353
e3e0520b
JB
1354 spin_lock(&block_group->lock);
1355 if (block_group->reserved || block_group->pinned ||
bf38be65 1356 block_group->used || block_group->ro ||
e3e0520b
JB
1357 list_is_singular(&block_group->list)) {
1358 /*
1359 * We want to bail if we made new allocations or have
1360 * outstanding allocations in this block group. We do
1361 * the ro check in case balance is currently acting on
1362 * this block group.
1363 */
1364 trace_btrfs_skip_unused_block_group(block_group);
1365 spin_unlock(&block_group->lock);
1366 up_write(&space_info->groups_sem);
1367 goto next;
1368 }
1369 spin_unlock(&block_group->lock);
1370
1371 /* We don't want to force the issue, only flip if it's ok. */
e11c0406 1372 ret = inc_block_group_ro(block_group, 0);
e3e0520b
JB
1373 up_write(&space_info->groups_sem);
1374 if (ret < 0) {
1375 ret = 0;
1376 goto next;
1377 }
1378
74e91b12
NA
1379 ret = btrfs_zone_finish(block_group);
1380 if (ret < 0) {
1381 btrfs_dec_block_group_ro(block_group);
1382 if (ret == -EAGAIN)
1383 ret = 0;
1384 goto next;
1385 }
1386
e3e0520b
JB
1387 /*
1388 * Want to do this before we do anything else so we can recover
1389 * properly if we fail to join the transaction.
1390 */
1391 trans = btrfs_start_trans_remove_block_group(fs_info,
b3470b5d 1392 block_group->start);
e3e0520b
JB
1393 if (IS_ERR(trans)) {
1394 btrfs_dec_block_group_ro(block_group);
1395 ret = PTR_ERR(trans);
1396 goto next;
1397 }
1398
1399 /*
1400 * We could have pending pinned extents for this block group,
1401 * just delete them, we don't care about them anymore.
1402 */
534cf531
FM
1403 if (!clean_pinned_extents(trans, block_group)) {
1404 btrfs_dec_block_group_ro(block_group);
e3e0520b 1405 goto end_trans;
534cf531 1406 }
e3e0520b 1407
b0643e59
DZ
1408 /*
1409 * At this point, the block_group is read only and should fail
1410 * new allocations. However, btrfs_finish_extent_commit() can
1411 * cause this block_group to be placed back on the discard
1412 * lists because now the block_group isn't fully discarded.
1413 * Bail here and try again later after discarding everything.
1414 */
1415 spin_lock(&fs_info->discard_ctl.lock);
1416 if (!list_empty(&block_group->discard_list)) {
1417 spin_unlock(&fs_info->discard_ctl.lock);
1418 btrfs_dec_block_group_ro(block_group);
1419 btrfs_discard_queue_work(&fs_info->discard_ctl,
1420 block_group);
1421 goto end_trans;
1422 }
1423 spin_unlock(&fs_info->discard_ctl.lock);
1424
e3e0520b
JB
1425 /* Reset pinned so btrfs_put_block_group doesn't complain */
1426 spin_lock(&space_info->lock);
1427 spin_lock(&block_group->lock);
1428
1429 btrfs_space_info_update_bytes_pinned(fs_info, space_info,
1430 -block_group->pinned);
1431 space_info->bytes_readonly += block_group->pinned;
e3e0520b
JB
1432 block_group->pinned = 0;
1433
1434 spin_unlock(&block_group->lock);
1435 spin_unlock(&space_info->lock);
1436
6e80d4f8
DZ
1437 /*
1438 * The normal path here is an unused block group is passed here,
1439 * then trimming is handled in the transaction commit path.
1440 * Async discard interposes before this to do the trimming
1441 * before coming down the unused block group path as trimming
1442 * will no longer be done later in the transaction commit path.
1443 */
1444 if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC))
1445 goto flip_async;
1446
dcba6e48
NA
1447 /*
1448 * DISCARD can flip during remount. On zoned filesystems, we
1449 * need to reset sequential-required zones.
1450 */
1451 trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) ||
1452 btrfs_is_zoned(fs_info);
e3e0520b
JB
1453
1454 /* Implicit trim during transaction commit. */
1455 if (trimming)
6b7304af 1456 btrfs_freeze_block_group(block_group);
e3e0520b
JB
1457
1458 /*
1459 * Btrfs_remove_chunk will abort the transaction if things go
1460 * horribly wrong.
1461 */
b3470b5d 1462 ret = btrfs_remove_chunk(trans, block_group->start);
e3e0520b
JB
1463
1464 if (ret) {
1465 if (trimming)
6b7304af 1466 btrfs_unfreeze_block_group(block_group);
e3e0520b
JB
1467 goto end_trans;
1468 }
1469
1470 /*
1471 * If we're not mounted with -odiscard, we can just forget
1472 * about this block group. Otherwise we'll need to wait
1473 * until transaction commit to do the actual discard.
1474 */
1475 if (trimming) {
1476 spin_lock(&fs_info->unused_bgs_lock);
1477 /*
1478 * A concurrent scrub might have added us to the list
1479 * fs_info->unused_bgs, so use a list_move operation
1480 * to add the block group to the deleted_bgs list.
1481 */
1482 list_move(&block_group->bg_list,
1483 &trans->transaction->deleted_bgs);
1484 spin_unlock(&fs_info->unused_bgs_lock);
1485 btrfs_get_block_group(block_group);
1486 }
1487end_trans:
1488 btrfs_end_transaction(trans);
1489next:
e3e0520b
JB
1490 btrfs_put_block_group(block_group);
1491 spin_lock(&fs_info->unused_bgs_lock);
1492 }
1493 spin_unlock(&fs_info->unused_bgs_lock);
f3372065 1494 mutex_unlock(&fs_info->reclaim_bgs_lock);
6e80d4f8
DZ
1495 return;
1496
1497flip_async:
1498 btrfs_end_transaction(trans);
f3372065 1499 mutex_unlock(&fs_info->reclaim_bgs_lock);
6e80d4f8
DZ
1500 btrfs_put_block_group(block_group);
1501 btrfs_discard_punt_unused_bgs_list(fs_info);
e3e0520b
JB
1502}
1503
32da5386 1504void btrfs_mark_bg_unused(struct btrfs_block_group *bg)
e3e0520b
JB
1505{
1506 struct btrfs_fs_info *fs_info = bg->fs_info;
1507
1508 spin_lock(&fs_info->unused_bgs_lock);
1509 if (list_empty(&bg->bg_list)) {
1510 btrfs_get_block_group(bg);
1511 trace_btrfs_add_unused_block_group(bg);
1512 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
1513 }
1514 spin_unlock(&fs_info->unused_bgs_lock);
1515}
4358d963 1516
2ca0ec77
JT
1517/*
1518 * We want block groups with a low number of used bytes to be in the beginning
1519 * of the list, so they will get reclaimed first.
1520 */
1521static int reclaim_bgs_cmp(void *unused, const struct list_head *a,
1522 const struct list_head *b)
1523{
1524 const struct btrfs_block_group *bg1, *bg2;
1525
1526 bg1 = list_entry(a, struct btrfs_block_group, bg_list);
1527 bg2 = list_entry(b, struct btrfs_block_group, bg_list);
1528
1529 return bg1->used > bg2->used;
1530}
1531
3687fcb0
JT
1532static inline bool btrfs_should_reclaim(struct btrfs_fs_info *fs_info)
1533{
1534 if (btrfs_is_zoned(fs_info))
1535 return btrfs_zoned_should_reclaim(fs_info);
1536 return true;
1537}
1538
18bb8bbf
JT
1539void btrfs_reclaim_bgs_work(struct work_struct *work)
1540{
1541 struct btrfs_fs_info *fs_info =
1542 container_of(work, struct btrfs_fs_info, reclaim_bgs_work);
1543 struct btrfs_block_group *bg;
1544 struct btrfs_space_info *space_info;
18bb8bbf
JT
1545
1546 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1547 return;
1548
2f12741f
JB
1549 if (btrfs_fs_closing(fs_info))
1550 return;
1551
3687fcb0
JT
1552 if (!btrfs_should_reclaim(fs_info))
1553 return;
1554
ca5e4ea0
NA
1555 sb_start_write(fs_info->sb);
1556
1557 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
1558 sb_end_write(fs_info->sb);
18bb8bbf 1559 return;
ca5e4ea0 1560 }
18bb8bbf 1561
9cc0b837
JT
1562 /*
1563 * Long running balances can keep us blocked here for eternity, so
1564 * simply skip reclaim if we're unable to get the mutex.
1565 */
1566 if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) {
1567 btrfs_exclop_finish(fs_info);
ca5e4ea0 1568 sb_end_write(fs_info->sb);
9cc0b837
JT
1569 return;
1570 }
1571
18bb8bbf 1572 spin_lock(&fs_info->unused_bgs_lock);
2ca0ec77
JT
1573 /*
1574 * Sort happens under lock because we can't simply splice it and sort.
1575 * The block groups might still be in use and reachable via bg_list,
1576 * and their presence in the reclaim_bgs list must be preserved.
1577 */
1578 list_sort(NULL, &fs_info->reclaim_bgs, reclaim_bgs_cmp);
18bb8bbf 1579 while (!list_empty(&fs_info->reclaim_bgs)) {
5f93e776 1580 u64 zone_unusable;
1cea5cf0
FM
1581 int ret = 0;
1582
18bb8bbf
JT
1583 bg = list_first_entry(&fs_info->reclaim_bgs,
1584 struct btrfs_block_group,
1585 bg_list);
1586 list_del_init(&bg->bg_list);
1587
1588 space_info = bg->space_info;
1589 spin_unlock(&fs_info->unused_bgs_lock);
1590
1591 /* Don't race with allocators so take the groups_sem */
1592 down_write(&space_info->groups_sem);
1593
1594 spin_lock(&bg->lock);
1595 if (bg->reserved || bg->pinned || bg->ro) {
1596 /*
1597 * We want to bail if we made new allocations or have
1598 * outstanding allocations in this block group. We do
1599 * the ro check in case balance is currently acting on
1600 * this block group.
1601 */
1602 spin_unlock(&bg->lock);
1603 up_write(&space_info->groups_sem);
1604 goto next;
1605 }
1606 spin_unlock(&bg->lock);
1607
1608 /* Get out fast, in case we're unmounting the filesystem */
1609 if (btrfs_fs_closing(fs_info)) {
1610 up_write(&space_info->groups_sem);
1611 goto next;
1612 }
1613
5f93e776
JT
1614 /*
1615 * Cache the zone_unusable value before turning the block group
1616 * to read only. As soon as the blog group is read only it's
1617 * zone_unusable value gets moved to the block group's read-only
1618 * bytes and isn't available for calculations anymore.
1619 */
1620 zone_unusable = bg->zone_unusable;
18bb8bbf
JT
1621 ret = inc_block_group_ro(bg, 0);
1622 up_write(&space_info->groups_sem);
1623 if (ret < 0)
1624 goto next;
1625
5f93e776
JT
1626 btrfs_info(fs_info,
1627 "reclaiming chunk %llu with %llu%% used %llu%% unusable",
1628 bg->start, div_u64(bg->used * 100, bg->length),
1629 div64_u64(zone_unusable * 100, bg->length));
18bb8bbf
JT
1630 trace_btrfs_reclaim_block_group(bg);
1631 ret = btrfs_relocate_chunk(fs_info, bg->start);
74944c87
JB
1632 if (ret) {
1633 btrfs_dec_block_group_ro(bg);
18bb8bbf
JT
1634 btrfs_err(fs_info, "error relocating chunk %llu",
1635 bg->start);
74944c87 1636 }
18bb8bbf
JT
1637
1638next:
d96b3424 1639 btrfs_put_block_group(bg);
18bb8bbf
JT
1640 spin_lock(&fs_info->unused_bgs_lock);
1641 }
1642 spin_unlock(&fs_info->unused_bgs_lock);
1643 mutex_unlock(&fs_info->reclaim_bgs_lock);
1644 btrfs_exclop_finish(fs_info);
ca5e4ea0 1645 sb_end_write(fs_info->sb);
18bb8bbf
JT
1646}
1647
1648void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info)
1649{
1650 spin_lock(&fs_info->unused_bgs_lock);
1651 if (!list_empty(&fs_info->reclaim_bgs))
1652 queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work);
1653 spin_unlock(&fs_info->unused_bgs_lock);
1654}
1655
1656void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg)
1657{
1658 struct btrfs_fs_info *fs_info = bg->fs_info;
1659
1660 spin_lock(&fs_info->unused_bgs_lock);
1661 if (list_empty(&bg->bg_list)) {
1662 btrfs_get_block_group(bg);
1663 trace_btrfs_add_reclaim_block_group(bg);
1664 list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs);
1665 }
1666 spin_unlock(&fs_info->unused_bgs_lock);
1667}
1668
e3ba67a1
JT
1669static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1670 struct btrfs_path *path)
1671{
1672 struct extent_map_tree *em_tree;
1673 struct extent_map *em;
1674 struct btrfs_block_group_item bg;
1675 struct extent_buffer *leaf;
1676 int slot;
1677 u64 flags;
1678 int ret = 0;
1679
1680 slot = path->slots[0];
1681 leaf = path->nodes[0];
1682
1683 em_tree = &fs_info->mapping_tree;
1684 read_lock(&em_tree->lock);
1685 em = lookup_extent_mapping(em_tree, key->objectid, key->offset);
1686 read_unlock(&em_tree->lock);
1687 if (!em) {
1688 btrfs_err(fs_info,
1689 "logical %llu len %llu found bg but no related chunk",
1690 key->objectid, key->offset);
1691 return -ENOENT;
1692 }
1693
1694 if (em->start != key->objectid || em->len != key->offset) {
1695 btrfs_err(fs_info,
1696 "block group %llu len %llu mismatch with chunk %llu len %llu",
1697 key->objectid, key->offset, em->start, em->len);
1698 ret = -EUCLEAN;
1699 goto out_free_em;
1700 }
1701
1702 read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot),
1703 sizeof(bg));
1704 flags = btrfs_stack_block_group_flags(&bg) &
1705 BTRFS_BLOCK_GROUP_TYPE_MASK;
1706
1707 if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1708 btrfs_err(fs_info,
1709"block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
1710 key->objectid, key->offset, flags,
1711 (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type));
1712 ret = -EUCLEAN;
1713 }
1714
1715out_free_em:
1716 free_extent_map(em);
1717 return ret;
1718}
1719
4358d963
JB
1720static int find_first_block_group(struct btrfs_fs_info *fs_info,
1721 struct btrfs_path *path,
1722 struct btrfs_key *key)
1723{
dfe8aec4 1724 struct btrfs_root *root = btrfs_block_group_root(fs_info);
e3ba67a1 1725 int ret;
4358d963 1726 struct btrfs_key found_key;
4358d963 1727
36dfbbe2 1728 btrfs_for_each_slot(root, key, &found_key, path, ret) {
4358d963
JB
1729 if (found_key.objectid >= key->objectid &&
1730 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
36dfbbe2 1731 return read_bg_from_eb(fs_info, &found_key, path);
4358d963 1732 }
4358d963 1733 }
4358d963
JB
1734 return ret;
1735}
1736
1737static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1738{
1739 u64 extra_flags = chunk_to_extended(flags) &
1740 BTRFS_EXTENDED_PROFILE_MASK;
1741
1742 write_seqlock(&fs_info->profiles_lock);
1743 if (flags & BTRFS_BLOCK_GROUP_DATA)
1744 fs_info->avail_data_alloc_bits |= extra_flags;
1745 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1746 fs_info->avail_metadata_alloc_bits |= extra_flags;
1747 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1748 fs_info->avail_system_alloc_bits |= extra_flags;
1749 write_sequnlock(&fs_info->profiles_lock);
1750}
1751
96a14336 1752/**
9ee9b979
NB
1753 * Map a physical disk address to a list of logical addresses
1754 *
1755 * @fs_info: the filesystem
96a14336 1756 * @chunk_start: logical address of block group
138082f3 1757 * @bdev: physical device to resolve, can be NULL to indicate any device
96a14336
NB
1758 * @physical: physical address to map to logical addresses
1759 * @logical: return array of logical addresses which map to @physical
1760 * @naddrs: length of @logical
1761 * @stripe_len: size of IO stripe for the given block group
1762 *
1763 * Maps a particular @physical disk address to a list of @logical addresses.
1764 * Used primarily to exclude those portions of a block group that contain super
1765 * block copies.
1766 */
96a14336 1767int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
138082f3
NA
1768 struct block_device *bdev, u64 physical, u64 **logical,
1769 int *naddrs, int *stripe_len)
96a14336
NB
1770{
1771 struct extent_map *em;
1772 struct map_lookup *map;
1773 u64 *buf;
1774 u64 bytenr;
1776ad17
NB
1775 u64 data_stripe_length;
1776 u64 io_stripe_size;
1777 int i, nr = 0;
1778 int ret = 0;
96a14336
NB
1779
1780 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
1781 if (IS_ERR(em))
1782 return -EIO;
1783
1784 map = em->map_lookup;
9e22b925 1785 data_stripe_length = em->orig_block_len;
1776ad17 1786 io_stripe_size = map->stripe_len;
138082f3 1787 chunk_start = em->start;
96a14336 1788
9e22b925
NB
1789 /* For RAID5/6 adjust to a full IO stripe length */
1790 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
1776ad17 1791 io_stripe_size = map->stripe_len * nr_data_stripes(map);
96a14336
NB
1792
1793 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
1776ad17
NB
1794 if (!buf) {
1795 ret = -ENOMEM;
1796 goto out;
1797 }
96a14336
NB
1798
1799 for (i = 0; i < map->num_stripes; i++) {
1776ad17
NB
1800 bool already_inserted = false;
1801 u64 stripe_nr;
138082f3 1802 u64 offset;
1776ad17
NB
1803 int j;
1804
1805 if (!in_range(physical, map->stripes[i].physical,
1806 data_stripe_length))
96a14336
NB
1807 continue;
1808
138082f3
NA
1809 if (bdev && map->stripes[i].dev->bdev != bdev)
1810 continue;
1811
96a14336 1812 stripe_nr = physical - map->stripes[i].physical;
138082f3 1813 stripe_nr = div64_u64_rem(stripe_nr, map->stripe_len, &offset);
96a14336 1814
ac067734
DS
1815 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
1816 BTRFS_BLOCK_GROUP_RAID10)) {
96a14336
NB
1817 stripe_nr = stripe_nr * map->num_stripes + i;
1818 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
96a14336
NB
1819 }
1820 /*
1821 * The remaining case would be for RAID56, multiply by
1822 * nr_data_stripes(). Alternatively, just use rmap_len below
1823 * instead of map->stripe_len
1824 */
1825
138082f3 1826 bytenr = chunk_start + stripe_nr * io_stripe_size + offset;
1776ad17
NB
1827
1828 /* Ensure we don't add duplicate addresses */
96a14336 1829 for (j = 0; j < nr; j++) {
1776ad17
NB
1830 if (buf[j] == bytenr) {
1831 already_inserted = true;
96a14336 1832 break;
1776ad17 1833 }
96a14336 1834 }
1776ad17
NB
1835
1836 if (!already_inserted)
96a14336 1837 buf[nr++] = bytenr;
96a14336
NB
1838 }
1839
1840 *logical = buf;
1841 *naddrs = nr;
1776ad17
NB
1842 *stripe_len = io_stripe_size;
1843out:
96a14336 1844 free_extent_map(em);
1776ad17 1845 return ret;
96a14336
NB
1846}
1847
32da5386 1848static int exclude_super_stripes(struct btrfs_block_group *cache)
4358d963
JB
1849{
1850 struct btrfs_fs_info *fs_info = cache->fs_info;
12659251 1851 const bool zoned = btrfs_is_zoned(fs_info);
4358d963
JB
1852 u64 bytenr;
1853 u64 *logical;
1854 int stripe_len;
1855 int i, nr, ret;
1856
b3470b5d
DS
1857 if (cache->start < BTRFS_SUPER_INFO_OFFSET) {
1858 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start;
4358d963 1859 cache->bytes_super += stripe_len;
b3470b5d 1860 ret = btrfs_add_excluded_extent(fs_info, cache->start,
4358d963
JB
1861 stripe_len);
1862 if (ret)
1863 return ret;
1864 }
1865
1866 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1867 bytenr = btrfs_sb_offset(i);
138082f3 1868 ret = btrfs_rmap_block(fs_info, cache->start, NULL,
4358d963
JB
1869 bytenr, &logical, &nr, &stripe_len);
1870 if (ret)
1871 return ret;
1872
12659251
NA
1873 /* Shouldn't have super stripes in sequential zones */
1874 if (zoned && nr) {
1875 btrfs_err(fs_info,
1876 "zoned: block group %llu must not contain super block",
1877 cache->start);
1878 return -EUCLEAN;
1879 }
1880
4358d963 1881 while (nr--) {
96f9b0f2
NB
1882 u64 len = min_t(u64, stripe_len,
1883 cache->start + cache->length - logical[nr]);
4358d963
JB
1884
1885 cache->bytes_super += len;
96f9b0f2
NB
1886 ret = btrfs_add_excluded_extent(fs_info, logical[nr],
1887 len);
4358d963
JB
1888 if (ret) {
1889 kfree(logical);
1890 return ret;
1891 }
1892 }
1893
1894 kfree(logical);
1895 }
1896 return 0;
1897}
1898
32da5386 1899static struct btrfs_block_group *btrfs_create_block_group_cache(
9afc6649 1900 struct btrfs_fs_info *fs_info, u64 start)
4358d963 1901{
32da5386 1902 struct btrfs_block_group *cache;
4358d963
JB
1903
1904 cache = kzalloc(sizeof(*cache), GFP_NOFS);
1905 if (!cache)
1906 return NULL;
1907
1908 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
1909 GFP_NOFS);
1910 if (!cache->free_space_ctl) {
1911 kfree(cache);
1912 return NULL;
1913 }
1914
b3470b5d 1915 cache->start = start;
4358d963
JB
1916
1917 cache->fs_info = fs_info;
1918 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
4358d963 1919
6e80d4f8
DZ
1920 cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED;
1921
48aaeebe 1922 refcount_set(&cache->refs, 1);
4358d963
JB
1923 spin_lock_init(&cache->lock);
1924 init_rwsem(&cache->data_rwsem);
1925 INIT_LIST_HEAD(&cache->list);
1926 INIT_LIST_HEAD(&cache->cluster_list);
1927 INIT_LIST_HEAD(&cache->bg_list);
1928 INIT_LIST_HEAD(&cache->ro_list);
b0643e59 1929 INIT_LIST_HEAD(&cache->discard_list);
4358d963
JB
1930 INIT_LIST_HEAD(&cache->dirty_list);
1931 INIT_LIST_HEAD(&cache->io_list);
afba2bc0 1932 INIT_LIST_HEAD(&cache->active_bg_list);
cd79909b 1933 btrfs_init_free_space_ctl(cache, cache->free_space_ctl);
6b7304af 1934 atomic_set(&cache->frozen, 0);
4358d963
JB
1935 mutex_init(&cache->free_space_lock);
1936 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
1937
1938 return cache;
1939}
1940
1941/*
1942 * Iterate all chunks and verify that each of them has the corresponding block
1943 * group
1944 */
1945static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
1946{
1947 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
1948 struct extent_map *em;
32da5386 1949 struct btrfs_block_group *bg;
4358d963
JB
1950 u64 start = 0;
1951 int ret = 0;
1952
1953 while (1) {
1954 read_lock(&map_tree->lock);
1955 /*
1956 * lookup_extent_mapping will return the first extent map
1957 * intersecting the range, so setting @len to 1 is enough to
1958 * get the first chunk.
1959 */
1960 em = lookup_extent_mapping(map_tree, start, 1);
1961 read_unlock(&map_tree->lock);
1962 if (!em)
1963 break;
1964
1965 bg = btrfs_lookup_block_group(fs_info, em->start);
1966 if (!bg) {
1967 btrfs_err(fs_info,
1968 "chunk start=%llu len=%llu doesn't have corresponding block group",
1969 em->start, em->len);
1970 ret = -EUCLEAN;
1971 free_extent_map(em);
1972 break;
1973 }
b3470b5d 1974 if (bg->start != em->start || bg->length != em->len ||
4358d963
JB
1975 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
1976 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1977 btrfs_err(fs_info,
1978"chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
1979 em->start, em->len,
1980 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
b3470b5d 1981 bg->start, bg->length,
4358d963
JB
1982 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
1983 ret = -EUCLEAN;
1984 free_extent_map(em);
1985 btrfs_put_block_group(bg);
1986 break;
1987 }
1988 start = em->start + em->len;
1989 free_extent_map(em);
1990 btrfs_put_block_group(bg);
1991 }
1992 return ret;
1993}
1994
ffb9e0f0 1995static int read_one_block_group(struct btrfs_fs_info *info,
4afd2fe8 1996 struct btrfs_block_group_item *bgi,
d49a2ddb 1997 const struct btrfs_key *key,
ffb9e0f0
QW
1998 int need_clear)
1999{
32da5386 2000 struct btrfs_block_group *cache;
ffb9e0f0 2001 const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS);
ffb9e0f0
QW
2002 int ret;
2003
d49a2ddb 2004 ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY);
ffb9e0f0 2005
9afc6649 2006 cache = btrfs_create_block_group_cache(info, key->objectid);
ffb9e0f0
QW
2007 if (!cache)
2008 return -ENOMEM;
2009
4afd2fe8
JT
2010 cache->length = key->offset;
2011 cache->used = btrfs_stack_block_group_used(bgi);
2012 cache->flags = btrfs_stack_block_group_flags(bgi);
f7238e50 2013 cache->global_root_id = btrfs_stack_block_group_chunk_objectid(bgi);
9afc6649 2014
e3e39c72
MPS
2015 set_free_space_tree_thresholds(cache);
2016
ffb9e0f0
QW
2017 if (need_clear) {
2018 /*
2019 * When we mount with old space cache, we need to
2020 * set BTRFS_DC_CLEAR and set dirty flag.
2021 *
2022 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
2023 * truncate the old free space cache inode and
2024 * setup a new one.
2025 * b) Setting 'dirty flag' makes sure that we flush
2026 * the new space cache info onto disk.
2027 */
2028 if (btrfs_test_opt(info, SPACE_CACHE))
2029 cache->disk_cache_state = BTRFS_DC_CLEAR;
2030 }
ffb9e0f0
QW
2031 if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
2032 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
2033 btrfs_err(info,
2034"bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
2035 cache->start);
2036 ret = -EINVAL;
2037 goto error;
2038 }
2039
a94794d5 2040 ret = btrfs_load_block_group_zone_info(cache, false);
08e11a3d
NA
2041 if (ret) {
2042 btrfs_err(info, "zoned: failed to load zone info of bg %llu",
2043 cache->start);
2044 goto error;
2045 }
2046
ffb9e0f0
QW
2047 /*
2048 * We need to exclude the super stripes now so that the space info has
2049 * super bytes accounted for, otherwise we'll think we have more space
2050 * than we actually do.
2051 */
2052 ret = exclude_super_stripes(cache);
2053 if (ret) {
2054 /* We may have excluded something, so call this just in case. */
2055 btrfs_free_excluded_extents(cache);
2056 goto error;
2057 }
2058
2059 /*
169e0da9
NA
2060 * For zoned filesystem, space after the allocation offset is the only
2061 * free space for a block group. So, we don't need any caching work.
2062 * btrfs_calc_zone_unusable() will set the amount of free space and
2063 * zone_unusable space.
2064 *
2065 * For regular filesystem, check for two cases, either we are full, and
2066 * therefore don't need to bother with the caching work since we won't
2067 * find any space, or we are empty, and we can just add all the space
2068 * in and be done with it. This saves us _a_lot_ of time, particularly
2069 * in the full case.
ffb9e0f0 2070 */
169e0da9
NA
2071 if (btrfs_is_zoned(info)) {
2072 btrfs_calc_zone_unusable(cache);
c46c4247
NA
2073 /* Should not have any excluded extents. Just in case, though. */
2074 btrfs_free_excluded_extents(cache);
169e0da9 2075 } else if (cache->length == cache->used) {
ffb9e0f0
QW
2076 cache->last_byte_to_unpin = (u64)-1;
2077 cache->cached = BTRFS_CACHE_FINISHED;
2078 btrfs_free_excluded_extents(cache);
2079 } else if (cache->used == 0) {
2080 cache->last_byte_to_unpin = (u64)-1;
2081 cache->cached = BTRFS_CACHE_FINISHED;
9afc6649
QW
2082 add_new_free_space(cache, cache->start,
2083 cache->start + cache->length);
ffb9e0f0
QW
2084 btrfs_free_excluded_extents(cache);
2085 }
2086
2087 ret = btrfs_add_block_group_cache(info, cache);
2088 if (ret) {
2089 btrfs_remove_free_space_cache(cache);
2090 goto error;
2091 }
2092 trace_btrfs_add_block_group(info, cache, 0);
723de71d 2093 btrfs_add_bg_to_space_info(info, cache);
ffb9e0f0
QW
2094
2095 set_avail_alloc_bits(info, cache->flags);
a09f23c3
AJ
2096 if (btrfs_chunk_writeable(info, cache->start)) {
2097 if (cache->used == 0) {
2098 ASSERT(list_empty(&cache->bg_list));
2099 if (btrfs_test_opt(info, DISCARD_ASYNC))
2100 btrfs_discard_queue_work(&info->discard_ctl, cache);
2101 else
2102 btrfs_mark_bg_unused(cache);
2103 }
2104 } else {
ffb9e0f0 2105 inc_block_group_ro(cache, 1);
ffb9e0f0 2106 }
a09f23c3 2107
ffb9e0f0
QW
2108 return 0;
2109error:
2110 btrfs_put_block_group(cache);
2111 return ret;
2112}
2113
42437a63
JB
2114static int fill_dummy_bgs(struct btrfs_fs_info *fs_info)
2115{
2116 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
42437a63
JB
2117 struct rb_node *node;
2118 int ret = 0;
2119
2120 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
2121 struct extent_map *em;
2122 struct map_lookup *map;
2123 struct btrfs_block_group *bg;
2124
2125 em = rb_entry(node, struct extent_map, rb_node);
2126 map = em->map_lookup;
2127 bg = btrfs_create_block_group_cache(fs_info, em->start);
2128 if (!bg) {
2129 ret = -ENOMEM;
2130 break;
2131 }
2132
2133 /* Fill dummy cache as FULL */
2134 bg->length = em->len;
2135 bg->flags = map->type;
2136 bg->last_byte_to_unpin = (u64)-1;
2137 bg->cached = BTRFS_CACHE_FINISHED;
2138 bg->used = em->len;
2139 bg->flags = map->type;
2140 ret = btrfs_add_block_group_cache(fs_info, bg);
2b29726c
QW
2141 /*
2142 * We may have some valid block group cache added already, in
2143 * that case we skip to the next one.
2144 */
2145 if (ret == -EEXIST) {
2146 ret = 0;
2147 btrfs_put_block_group(bg);
2148 continue;
2149 }
2150
42437a63
JB
2151 if (ret) {
2152 btrfs_remove_free_space_cache(bg);
2153 btrfs_put_block_group(bg);
2154 break;
2155 }
2b29726c 2156
723de71d 2157 btrfs_add_bg_to_space_info(fs_info, bg);
42437a63
JB
2158
2159 set_avail_alloc_bits(fs_info, bg->flags);
2160 }
2161 if (!ret)
2162 btrfs_init_global_block_rsv(fs_info);
2163 return ret;
2164}
2165
4358d963
JB
2166int btrfs_read_block_groups(struct btrfs_fs_info *info)
2167{
dfe8aec4 2168 struct btrfs_root *root = btrfs_block_group_root(info);
4358d963
JB
2169 struct btrfs_path *path;
2170 int ret;
32da5386 2171 struct btrfs_block_group *cache;
4358d963
JB
2172 struct btrfs_space_info *space_info;
2173 struct btrfs_key key;
4358d963
JB
2174 int need_clear = 0;
2175 u64 cache_gen;
4358d963 2176
dfe8aec4 2177 if (!root)
42437a63
JB
2178 return fill_dummy_bgs(info);
2179
4358d963
JB
2180 key.objectid = 0;
2181 key.offset = 0;
2182 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2183 path = btrfs_alloc_path();
2184 if (!path)
2185 return -ENOMEM;
4358d963
JB
2186
2187 cache_gen = btrfs_super_cache_generation(info->super_copy);
2188 if (btrfs_test_opt(info, SPACE_CACHE) &&
2189 btrfs_super_generation(info->super_copy) != cache_gen)
2190 need_clear = 1;
2191 if (btrfs_test_opt(info, CLEAR_CACHE))
2192 need_clear = 1;
2193
2194 while (1) {
4afd2fe8
JT
2195 struct btrfs_block_group_item bgi;
2196 struct extent_buffer *leaf;
2197 int slot;
2198
4358d963
JB
2199 ret = find_first_block_group(info, path, &key);
2200 if (ret > 0)
2201 break;
2202 if (ret != 0)
2203 goto error;
2204
4afd2fe8
JT
2205 leaf = path->nodes[0];
2206 slot = path->slots[0];
2207
2208 read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
2209 sizeof(bgi));
2210
2211 btrfs_item_key_to_cpu(leaf, &key, slot);
2212 btrfs_release_path(path);
2213 ret = read_one_block_group(info, &bgi, &key, need_clear);
ffb9e0f0 2214 if (ret < 0)
4358d963 2215 goto error;
ffb9e0f0
QW
2216 key.objectid += key.offset;
2217 key.offset = 0;
4358d963 2218 }
7837fa88 2219 btrfs_release_path(path);
4358d963 2220
72804905 2221 list_for_each_entry(space_info, &info->space_info, list) {
49ea112d
JB
2222 int i;
2223
2224 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2225 if (list_empty(&space_info->block_groups[i]))
2226 continue;
2227 cache = list_first_entry(&space_info->block_groups[i],
2228 struct btrfs_block_group,
2229 list);
2230 btrfs_sysfs_add_block_group_type(cache);
2231 }
2232
4358d963
JB
2233 if (!(btrfs_get_alloc_profile(info, space_info->flags) &
2234 (BTRFS_BLOCK_GROUP_RAID10 |
2235 BTRFS_BLOCK_GROUP_RAID1_MASK |
2236 BTRFS_BLOCK_GROUP_RAID56_MASK |
2237 BTRFS_BLOCK_GROUP_DUP)))
2238 continue;
2239 /*
2240 * Avoid allocating from un-mirrored block group if there are
2241 * mirrored block groups.
2242 */
2243 list_for_each_entry(cache,
2244 &space_info->block_groups[BTRFS_RAID_RAID0],
2245 list)
e11c0406 2246 inc_block_group_ro(cache, 1);
4358d963
JB
2247 list_for_each_entry(cache,
2248 &space_info->block_groups[BTRFS_RAID_SINGLE],
2249 list)
e11c0406 2250 inc_block_group_ro(cache, 1);
4358d963
JB
2251 }
2252
2253 btrfs_init_global_block_rsv(info);
2254 ret = check_chunk_block_group_mappings(info);
2255error:
2256 btrfs_free_path(path);
2b29726c
QW
2257 /*
2258 * We've hit some error while reading the extent tree, and have
2259 * rescue=ibadroots mount option.
2260 * Try to fill the tree using dummy block groups so that the user can
2261 * continue to mount and grab their data.
2262 */
2263 if (ret && btrfs_test_opt(info, IGNOREBADROOTS))
2264 ret = fill_dummy_bgs(info);
4358d963
JB
2265 return ret;
2266}
2267
79bd3712
FM
2268/*
2269 * This function, insert_block_group_item(), belongs to the phase 2 of chunk
2270 * allocation.
2271 *
2272 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2273 * phases.
2274 */
97f4728a
QW
2275static int insert_block_group_item(struct btrfs_trans_handle *trans,
2276 struct btrfs_block_group *block_group)
2277{
2278 struct btrfs_fs_info *fs_info = trans->fs_info;
2279 struct btrfs_block_group_item bgi;
dfe8aec4 2280 struct btrfs_root *root = btrfs_block_group_root(fs_info);
97f4728a
QW
2281 struct btrfs_key key;
2282
2283 spin_lock(&block_group->lock);
2284 btrfs_set_stack_block_group_used(&bgi, block_group->used);
2285 btrfs_set_stack_block_group_chunk_objectid(&bgi,
f7238e50 2286 block_group->global_root_id);
97f4728a
QW
2287 btrfs_set_stack_block_group_flags(&bgi, block_group->flags);
2288 key.objectid = block_group->start;
2289 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2290 key.offset = block_group->length;
2291 spin_unlock(&block_group->lock);
2292
97f4728a
QW
2293 return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi));
2294}
2295
2eadb9e7
NB
2296static int insert_dev_extent(struct btrfs_trans_handle *trans,
2297 struct btrfs_device *device, u64 chunk_offset,
2298 u64 start, u64 num_bytes)
2299{
2300 struct btrfs_fs_info *fs_info = device->fs_info;
2301 struct btrfs_root *root = fs_info->dev_root;
2302 struct btrfs_path *path;
2303 struct btrfs_dev_extent *extent;
2304 struct extent_buffer *leaf;
2305 struct btrfs_key key;
2306 int ret;
2307
2308 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
2309 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
2310 path = btrfs_alloc_path();
2311 if (!path)
2312 return -ENOMEM;
2313
2314 key.objectid = device->devid;
2315 key.type = BTRFS_DEV_EXTENT_KEY;
2316 key.offset = start;
2317 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*extent));
2318 if (ret)
2319 goto out;
2320
2321 leaf = path->nodes[0];
2322 extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
2323 btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID);
2324 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
2325 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
2326 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
2327
2328 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
2329 btrfs_mark_buffer_dirty(leaf);
2330out:
2331 btrfs_free_path(path);
2332 return ret;
2333}
2334
2335/*
2336 * This function belongs to phase 2.
2337 *
2338 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2339 * phases.
2340 */
2341static int insert_dev_extents(struct btrfs_trans_handle *trans,
2342 u64 chunk_offset, u64 chunk_size)
2343{
2344 struct btrfs_fs_info *fs_info = trans->fs_info;
2345 struct btrfs_device *device;
2346 struct extent_map *em;
2347 struct map_lookup *map;
2348 u64 dev_offset;
2349 u64 stripe_size;
2350 int i;
2351 int ret = 0;
2352
2353 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
2354 if (IS_ERR(em))
2355 return PTR_ERR(em);
2356
2357 map = em->map_lookup;
2358 stripe_size = em->orig_block_len;
2359
2360 /*
2361 * Take the device list mutex to prevent races with the final phase of
2362 * a device replace operation that replaces the device object associated
2363 * with the map's stripes, because the device object's id can change
2364 * at any time during that final phase of the device replace operation
2365 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
2366 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
2367 * resulting in persisting a device extent item with such ID.
2368 */
2369 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2370 for (i = 0; i < map->num_stripes; i++) {
2371 device = map->stripes[i].dev;
2372 dev_offset = map->stripes[i].physical;
2373
2374 ret = insert_dev_extent(trans, device, chunk_offset, dev_offset,
2375 stripe_size);
2376 if (ret)
2377 break;
2378 }
2379 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2380
2381 free_extent_map(em);
2382 return ret;
2383}
2384
79bd3712
FM
2385/*
2386 * This function, btrfs_create_pending_block_groups(), belongs to the phase 2 of
2387 * chunk allocation.
2388 *
2389 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2390 * phases.
2391 */
4358d963
JB
2392void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
2393{
2394 struct btrfs_fs_info *fs_info = trans->fs_info;
32da5386 2395 struct btrfs_block_group *block_group;
4358d963
JB
2396 int ret = 0;
2397
4358d963 2398 while (!list_empty(&trans->new_bgs)) {
49ea112d
JB
2399 int index;
2400
4358d963 2401 block_group = list_first_entry(&trans->new_bgs,
32da5386 2402 struct btrfs_block_group,
4358d963
JB
2403 bg_list);
2404 if (ret)
2405 goto next;
2406
49ea112d
JB
2407 index = btrfs_bg_flags_to_raid_index(block_group->flags);
2408
97f4728a 2409 ret = insert_block_group_item(trans, block_group);
4358d963
JB
2410 if (ret)
2411 btrfs_abort_transaction(trans, ret);
79bd3712
FM
2412 if (!block_group->chunk_item_inserted) {
2413 mutex_lock(&fs_info->chunk_mutex);
2414 ret = btrfs_chunk_alloc_add_chunk_item(trans, block_group);
2415 mutex_unlock(&fs_info->chunk_mutex);
2416 if (ret)
2417 btrfs_abort_transaction(trans, ret);
2418 }
2eadb9e7
NB
2419 ret = insert_dev_extents(trans, block_group->start,
2420 block_group->length);
4358d963
JB
2421 if (ret)
2422 btrfs_abort_transaction(trans, ret);
2423 add_block_group_free_space(trans, block_group);
49ea112d
JB
2424
2425 /*
2426 * If we restriped during balance, we may have added a new raid
2427 * type, so now add the sysfs entries when it is safe to do so.
2428 * We don't have to worry about locking here as it's handled in
2429 * btrfs_sysfs_add_block_group_type.
2430 */
2431 if (block_group->space_info->block_group_kobjs[index] == NULL)
2432 btrfs_sysfs_add_block_group_type(block_group);
2433
4358d963
JB
2434 /* Already aborted the transaction if it failed. */
2435next:
2436 btrfs_delayed_refs_rsv_release(fs_info, 1);
2437 list_del_init(&block_group->bg_list);
2438 }
2439 btrfs_trans_release_chunk_metadata(trans);
2440}
2441
f7238e50
JB
2442/*
2443 * For extent tree v2 we use the block_group_item->chunk_offset to point at our
2444 * global root id. For v1 it's always set to BTRFS_FIRST_CHUNK_TREE_OBJECTID.
2445 */
2446static u64 calculate_global_root_id(struct btrfs_fs_info *fs_info, u64 offset)
2447{
2448 u64 div = SZ_1G;
2449 u64 index;
2450
2451 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2452 return BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2453
2454 /* If we have a smaller fs index based on 128MiB. */
2455 if (btrfs_super_total_bytes(fs_info->super_copy) <= (SZ_1G * 10ULL))
2456 div = SZ_128M;
2457
2458 offset = div64_u64(offset, div);
2459 div64_u64_rem(offset, fs_info->nr_global_roots, &index);
2460 return index;
2461}
2462
79bd3712
FM
2463struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *trans,
2464 u64 bytes_used, u64 type,
2465 u64 chunk_offset, u64 size)
4358d963
JB
2466{
2467 struct btrfs_fs_info *fs_info = trans->fs_info;
32da5386 2468 struct btrfs_block_group *cache;
4358d963
JB
2469 int ret;
2470
2471 btrfs_set_log_full_commit(trans);
2472
9afc6649 2473 cache = btrfs_create_block_group_cache(fs_info, chunk_offset);
4358d963 2474 if (!cache)
79bd3712 2475 return ERR_PTR(-ENOMEM);
4358d963 2476
9afc6649 2477 cache->length = size;
e3e39c72 2478 set_free_space_tree_thresholds(cache);
bf38be65 2479 cache->used = bytes_used;
4358d963
JB
2480 cache->flags = type;
2481 cache->last_byte_to_unpin = (u64)-1;
2482 cache->cached = BTRFS_CACHE_FINISHED;
f7238e50
JB
2483 cache->global_root_id = calculate_global_root_id(fs_info, cache->start);
2484
997e3e2e
BB
2485 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
2486 cache->needs_free_space = 1;
08e11a3d 2487
a94794d5 2488 ret = btrfs_load_block_group_zone_info(cache, true);
08e11a3d
NA
2489 if (ret) {
2490 btrfs_put_block_group(cache);
79bd3712 2491 return ERR_PTR(ret);
08e11a3d
NA
2492 }
2493
4358d963
JB
2494 ret = exclude_super_stripes(cache);
2495 if (ret) {
2496 /* We may have excluded something, so call this just in case */
2497 btrfs_free_excluded_extents(cache);
2498 btrfs_put_block_group(cache);
79bd3712 2499 return ERR_PTR(ret);
4358d963
JB
2500 }
2501
2502 add_new_free_space(cache, chunk_offset, chunk_offset + size);
2503
2504 btrfs_free_excluded_extents(cache);
2505
4358d963
JB
2506 /*
2507 * Ensure the corresponding space_info object is created and
2508 * assigned to our block group. We want our bg to be added to the rbtree
2509 * with its ->space_info set.
2510 */
2511 cache->space_info = btrfs_find_space_info(fs_info, cache->flags);
2512 ASSERT(cache->space_info);
2513
2514 ret = btrfs_add_block_group_cache(fs_info, cache);
2515 if (ret) {
2516 btrfs_remove_free_space_cache(cache);
2517 btrfs_put_block_group(cache);
79bd3712 2518 return ERR_PTR(ret);
4358d963
JB
2519 }
2520
2521 /*
2522 * Now that our block group has its ->space_info set and is inserted in
2523 * the rbtree, update the space info's counters.
2524 */
2525 trace_btrfs_add_block_group(fs_info, cache, 1);
723de71d 2526 btrfs_add_bg_to_space_info(fs_info, cache);
4358d963
JB
2527 btrfs_update_global_block_rsv(fs_info);
2528
9d4b0a12
JB
2529#ifdef CONFIG_BTRFS_DEBUG
2530 if (btrfs_should_fragment_free_space(cache)) {
2531 u64 new_bytes_used = size - bytes_used;
2532
2533 cache->space_info->bytes_used += new_bytes_used >> 1;
2534 fragment_free_space(cache);
2535 }
2536#endif
4358d963
JB
2537
2538 list_add_tail(&cache->bg_list, &trans->new_bgs);
2539 trans->delayed_ref_updates++;
2540 btrfs_update_delayed_refs_rsv(trans);
2541
2542 set_avail_alloc_bits(fs_info, type);
79bd3712 2543 return cache;
4358d963 2544}
26ce2095 2545
b12de528
QW
2546/*
2547 * Mark one block group RO, can be called several times for the same block
2548 * group.
2549 *
2550 * @cache: the destination block group
2551 * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to
2552 * ensure we still have some free space after marking this
2553 * block group RO.
2554 */
2555int btrfs_inc_block_group_ro(struct btrfs_block_group *cache,
2556 bool do_chunk_alloc)
26ce2095
JB
2557{
2558 struct btrfs_fs_info *fs_info = cache->fs_info;
2559 struct btrfs_trans_handle *trans;
dfe8aec4 2560 struct btrfs_root *root = btrfs_block_group_root(fs_info);
26ce2095
JB
2561 u64 alloc_flags;
2562 int ret;
b6e9f16c 2563 bool dirty_bg_running;
26ce2095 2564
2d192fc4
QW
2565 /*
2566 * This can only happen when we are doing read-only scrub on read-only
2567 * mount.
2568 * In that case we should not start a new transaction on read-only fs.
2569 * Thus here we skip all chunk allocations.
2570 */
2571 if (sb_rdonly(fs_info->sb)) {
2572 mutex_lock(&fs_info->ro_block_group_mutex);
2573 ret = inc_block_group_ro(cache, 0);
2574 mutex_unlock(&fs_info->ro_block_group_mutex);
2575 return ret;
2576 }
2577
b6e9f16c 2578 do {
dfe8aec4 2579 trans = btrfs_join_transaction(root);
b6e9f16c
NB
2580 if (IS_ERR(trans))
2581 return PTR_ERR(trans);
26ce2095 2582
b6e9f16c 2583 dirty_bg_running = false;
26ce2095 2584
b6e9f16c
NB
2585 /*
2586 * We're not allowed to set block groups readonly after the dirty
2587 * block group cache has started writing. If it already started,
2588 * back off and let this transaction commit.
2589 */
2590 mutex_lock(&fs_info->ro_block_group_mutex);
2591 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
2592 u64 transid = trans->transid;
26ce2095 2593
b6e9f16c
NB
2594 mutex_unlock(&fs_info->ro_block_group_mutex);
2595 btrfs_end_transaction(trans);
2596
2597 ret = btrfs_wait_for_commit(fs_info, transid);
2598 if (ret)
2599 return ret;
2600 dirty_bg_running = true;
2601 }
2602 } while (dirty_bg_running);
26ce2095 2603
b12de528 2604 if (do_chunk_alloc) {
26ce2095 2605 /*
b12de528
QW
2606 * If we are changing raid levels, try to allocate a
2607 * corresponding block group with the new raid level.
26ce2095 2608 */
349e120e 2609 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags);
b12de528
QW
2610 if (alloc_flags != cache->flags) {
2611 ret = btrfs_chunk_alloc(trans, alloc_flags,
2612 CHUNK_ALLOC_FORCE);
2613 /*
2614 * ENOSPC is allowed here, we may have enough space
2615 * already allocated at the new raid level to carry on
2616 */
2617 if (ret == -ENOSPC)
2618 ret = 0;
2619 if (ret < 0)
2620 goto out;
2621 }
26ce2095
JB
2622 }
2623
a7a63acc 2624 ret = inc_block_group_ro(cache, 0);
195a49ea 2625 if (!do_chunk_alloc || ret == -ETXTBSY)
b12de528 2626 goto unlock_out;
26ce2095
JB
2627 if (!ret)
2628 goto out;
2629 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags);
2630 ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
2631 if (ret < 0)
2632 goto out;
b6a98021
NA
2633 /*
2634 * We have allocated a new chunk. We also need to activate that chunk to
2635 * grant metadata tickets for zoned filesystem.
2636 */
2637 ret = btrfs_zoned_activate_one_bg(fs_info, cache->space_info, true);
2638 if (ret < 0)
2639 goto out;
2640
e11c0406 2641 ret = inc_block_group_ro(cache, 0);
195a49ea
FM
2642 if (ret == -ETXTBSY)
2643 goto unlock_out;
26ce2095
JB
2644out:
2645 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
349e120e 2646 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags);
26ce2095
JB
2647 mutex_lock(&fs_info->chunk_mutex);
2648 check_system_chunk(trans, alloc_flags);
2649 mutex_unlock(&fs_info->chunk_mutex);
2650 }
b12de528 2651unlock_out:
26ce2095
JB
2652 mutex_unlock(&fs_info->ro_block_group_mutex);
2653
2654 btrfs_end_transaction(trans);
2655 return ret;
2656}
2657
32da5386 2658void btrfs_dec_block_group_ro(struct btrfs_block_group *cache)
26ce2095
JB
2659{
2660 struct btrfs_space_info *sinfo = cache->space_info;
2661 u64 num_bytes;
2662
2663 BUG_ON(!cache->ro);
2664
2665 spin_lock(&sinfo->lock);
2666 spin_lock(&cache->lock);
2667 if (!--cache->ro) {
169e0da9
NA
2668 if (btrfs_is_zoned(cache->fs_info)) {
2669 /* Migrate zone_unusable bytes back */
98173255
NA
2670 cache->zone_unusable =
2671 (cache->alloc_offset - cache->used) +
2672 (cache->length - cache->zone_capacity);
169e0da9
NA
2673 sinfo->bytes_zone_unusable += cache->zone_unusable;
2674 sinfo->bytes_readonly -= cache->zone_unusable;
2675 }
f9f28e5b
NA
2676 num_bytes = cache->length - cache->reserved -
2677 cache->pinned - cache->bytes_super -
2678 cache->zone_unusable - cache->used;
2679 sinfo->bytes_readonly -= num_bytes;
26ce2095
JB
2680 list_del_init(&cache->ro_list);
2681 }
2682 spin_unlock(&cache->lock);
2683 spin_unlock(&sinfo->lock);
2684}
77745c05 2685
3be4d8ef
QW
2686static int update_block_group_item(struct btrfs_trans_handle *trans,
2687 struct btrfs_path *path,
2688 struct btrfs_block_group *cache)
77745c05
JB
2689{
2690 struct btrfs_fs_info *fs_info = trans->fs_info;
2691 int ret;
dfe8aec4 2692 struct btrfs_root *root = btrfs_block_group_root(fs_info);
77745c05
JB
2693 unsigned long bi;
2694 struct extent_buffer *leaf;
bf38be65 2695 struct btrfs_block_group_item bgi;
b3470b5d
DS
2696 struct btrfs_key key;
2697
2698 key.objectid = cache->start;
2699 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2700 key.offset = cache->length;
77745c05 2701
3be4d8ef 2702 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
77745c05
JB
2703 if (ret) {
2704 if (ret > 0)
2705 ret = -ENOENT;
2706 goto fail;
2707 }
2708
2709 leaf = path->nodes[0];
2710 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
de0dc456
DS
2711 btrfs_set_stack_block_group_used(&bgi, cache->used);
2712 btrfs_set_stack_block_group_chunk_objectid(&bgi,
f7238e50 2713 cache->global_root_id);
de0dc456 2714 btrfs_set_stack_block_group_flags(&bgi, cache->flags);
bf38be65 2715 write_extent_buffer(leaf, &bgi, bi, sizeof(bgi));
77745c05
JB
2716 btrfs_mark_buffer_dirty(leaf);
2717fail:
2718 btrfs_release_path(path);
2719 return ret;
2720
2721}
2722
32da5386 2723static int cache_save_setup(struct btrfs_block_group *block_group,
77745c05
JB
2724 struct btrfs_trans_handle *trans,
2725 struct btrfs_path *path)
2726{
2727 struct btrfs_fs_info *fs_info = block_group->fs_info;
2728 struct btrfs_root *root = fs_info->tree_root;
2729 struct inode *inode = NULL;
2730 struct extent_changeset *data_reserved = NULL;
2731 u64 alloc_hint = 0;
2732 int dcs = BTRFS_DC_ERROR;
0044ae11 2733 u64 cache_size = 0;
77745c05
JB
2734 int retries = 0;
2735 int ret = 0;
2736
af456a2c
BB
2737 if (!btrfs_test_opt(fs_info, SPACE_CACHE))
2738 return 0;
2739
77745c05
JB
2740 /*
2741 * If this block group is smaller than 100 megs don't bother caching the
2742 * block group.
2743 */
b3470b5d 2744 if (block_group->length < (100 * SZ_1M)) {
77745c05
JB
2745 spin_lock(&block_group->lock);
2746 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2747 spin_unlock(&block_group->lock);
2748 return 0;
2749 }
2750
bf31f87f 2751 if (TRANS_ABORTED(trans))
77745c05
JB
2752 return 0;
2753again:
2754 inode = lookup_free_space_inode(block_group, path);
2755 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2756 ret = PTR_ERR(inode);
2757 btrfs_release_path(path);
2758 goto out;
2759 }
2760
2761 if (IS_ERR(inode)) {
2762 BUG_ON(retries);
2763 retries++;
2764
2765 if (block_group->ro)
2766 goto out_free;
2767
2768 ret = create_free_space_inode(trans, block_group, path);
2769 if (ret)
2770 goto out_free;
2771 goto again;
2772 }
2773
2774 /*
2775 * We want to set the generation to 0, that way if anything goes wrong
2776 * from here on out we know not to trust this cache when we load up next
2777 * time.
2778 */
2779 BTRFS_I(inode)->generation = 0;
9a56fcd1 2780 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
77745c05
JB
2781 if (ret) {
2782 /*
2783 * So theoretically we could recover from this, simply set the
2784 * super cache generation to 0 so we know to invalidate the
2785 * cache, but then we'd have to keep track of the block groups
2786 * that fail this way so we know we _have_ to reset this cache
2787 * before the next commit or risk reading stale cache. So to
2788 * limit our exposure to horrible edge cases lets just abort the
2789 * transaction, this only happens in really bad situations
2790 * anyway.
2791 */
2792 btrfs_abort_transaction(trans, ret);
2793 goto out_put;
2794 }
2795 WARN_ON(ret);
2796
2797 /* We've already setup this transaction, go ahead and exit */
2798 if (block_group->cache_generation == trans->transid &&
2799 i_size_read(inode)) {
2800 dcs = BTRFS_DC_SETUP;
2801 goto out_put;
2802 }
2803
2804 if (i_size_read(inode) > 0) {
2805 ret = btrfs_check_trunc_cache_free_space(fs_info,
2806 &fs_info->global_block_rsv);
2807 if (ret)
2808 goto out_put;
2809
2810 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
2811 if (ret)
2812 goto out_put;
2813 }
2814
2815 spin_lock(&block_group->lock);
2816 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2817 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
2818 /*
2819 * don't bother trying to write stuff out _if_
2820 * a) we're not cached,
2821 * b) we're with nospace_cache mount option,
2822 * c) we're with v2 space_cache (FREE_SPACE_TREE).
2823 */
2824 dcs = BTRFS_DC_WRITTEN;
2825 spin_unlock(&block_group->lock);
2826 goto out_put;
2827 }
2828 spin_unlock(&block_group->lock);
2829
2830 /*
2831 * We hit an ENOSPC when setting up the cache in this transaction, just
2832 * skip doing the setup, we've already cleared the cache so we're safe.
2833 */
2834 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
2835 ret = -ENOSPC;
2836 goto out_put;
2837 }
2838
2839 /*
2840 * Try to preallocate enough space based on how big the block group is.
2841 * Keep in mind this has to include any pinned space which could end up
2842 * taking up quite a bit since it's not folded into the other space
2843 * cache.
2844 */
0044ae11
QW
2845 cache_size = div_u64(block_group->length, SZ_256M);
2846 if (!cache_size)
2847 cache_size = 1;
77745c05 2848
0044ae11
QW
2849 cache_size *= 16;
2850 cache_size *= fs_info->sectorsize;
77745c05 2851
36ea6f3e 2852 ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0,
0044ae11 2853 cache_size);
77745c05
JB
2854 if (ret)
2855 goto out_put;
2856
0044ae11
QW
2857 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, cache_size,
2858 cache_size, cache_size,
77745c05
JB
2859 &alloc_hint);
2860 /*
2861 * Our cache requires contiguous chunks so that we don't modify a bunch
2862 * of metadata or split extents when writing the cache out, which means
2863 * we can enospc if we are heavily fragmented in addition to just normal
2864 * out of space conditions. So if we hit this just skip setting up any
2865 * other block groups for this transaction, maybe we'll unpin enough
2866 * space the next time around.
2867 */
2868 if (!ret)
2869 dcs = BTRFS_DC_SETUP;
2870 else if (ret == -ENOSPC)
2871 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
2872
2873out_put:
2874 iput(inode);
2875out_free:
2876 btrfs_release_path(path);
2877out:
2878 spin_lock(&block_group->lock);
2879 if (!ret && dcs == BTRFS_DC_SETUP)
2880 block_group->cache_generation = trans->transid;
2881 block_group->disk_cache_state = dcs;
2882 spin_unlock(&block_group->lock);
2883
2884 extent_changeset_free(data_reserved);
2885 return ret;
2886}
2887
2888int btrfs_setup_space_cache(struct btrfs_trans_handle *trans)
2889{
2890 struct btrfs_fs_info *fs_info = trans->fs_info;
32da5386 2891 struct btrfs_block_group *cache, *tmp;
77745c05
JB
2892 struct btrfs_transaction *cur_trans = trans->transaction;
2893 struct btrfs_path *path;
2894
2895 if (list_empty(&cur_trans->dirty_bgs) ||
2896 !btrfs_test_opt(fs_info, SPACE_CACHE))
2897 return 0;
2898
2899 path = btrfs_alloc_path();
2900 if (!path)
2901 return -ENOMEM;
2902
2903 /* Could add new block groups, use _safe just in case */
2904 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
2905 dirty_list) {
2906 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2907 cache_save_setup(cache, trans, path);
2908 }
2909
2910 btrfs_free_path(path);
2911 return 0;
2912}
2913
2914/*
2915 * Transaction commit does final block group cache writeback during a critical
2916 * section where nothing is allowed to change the FS. This is required in
2917 * order for the cache to actually match the block group, but can introduce a
2918 * lot of latency into the commit.
2919 *
2920 * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO.
2921 * There's a chance we'll have to redo some of it if the block group changes
2922 * again during the commit, but it greatly reduces the commit latency by
2923 * getting rid of the easy block groups while we're still allowing others to
2924 * join the commit.
2925 */
2926int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
2927{
2928 struct btrfs_fs_info *fs_info = trans->fs_info;
32da5386 2929 struct btrfs_block_group *cache;
77745c05
JB
2930 struct btrfs_transaction *cur_trans = trans->transaction;
2931 int ret = 0;
2932 int should_put;
2933 struct btrfs_path *path = NULL;
2934 LIST_HEAD(dirty);
2935 struct list_head *io = &cur_trans->io_bgs;
77745c05
JB
2936 int loops = 0;
2937
2938 spin_lock(&cur_trans->dirty_bgs_lock);
2939 if (list_empty(&cur_trans->dirty_bgs)) {
2940 spin_unlock(&cur_trans->dirty_bgs_lock);
2941 return 0;
2942 }
2943 list_splice_init(&cur_trans->dirty_bgs, &dirty);
2944 spin_unlock(&cur_trans->dirty_bgs_lock);
2945
2946again:
2947 /* Make sure all the block groups on our dirty list actually exist */
2948 btrfs_create_pending_block_groups(trans);
2949
2950 if (!path) {
2951 path = btrfs_alloc_path();
938fcbfb
JB
2952 if (!path) {
2953 ret = -ENOMEM;
2954 goto out;
2955 }
77745c05
JB
2956 }
2957
2958 /*
2959 * cache_write_mutex is here only to save us from balance or automatic
2960 * removal of empty block groups deleting this block group while we are
2961 * writing out the cache
2962 */
2963 mutex_lock(&trans->transaction->cache_write_mutex);
2964 while (!list_empty(&dirty)) {
2965 bool drop_reserve = true;
2966
32da5386 2967 cache = list_first_entry(&dirty, struct btrfs_block_group,
77745c05
JB
2968 dirty_list);
2969 /*
2970 * This can happen if something re-dirties a block group that
2971 * is already under IO. Just wait for it to finish and then do
2972 * it all again
2973 */
2974 if (!list_empty(&cache->io_list)) {
2975 list_del_init(&cache->io_list);
2976 btrfs_wait_cache_io(trans, cache, path);
2977 btrfs_put_block_group(cache);
2978 }
2979
2980
2981 /*
2982 * btrfs_wait_cache_io uses the cache->dirty_list to decide if
2983 * it should update the cache_state. Don't delete until after
2984 * we wait.
2985 *
2986 * Since we're not running in the commit critical section
2987 * we need the dirty_bgs_lock to protect from update_block_group
2988 */
2989 spin_lock(&cur_trans->dirty_bgs_lock);
2990 list_del_init(&cache->dirty_list);
2991 spin_unlock(&cur_trans->dirty_bgs_lock);
2992
2993 should_put = 1;
2994
2995 cache_save_setup(cache, trans, path);
2996
2997 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
2998 cache->io_ctl.inode = NULL;
2999 ret = btrfs_write_out_cache(trans, cache, path);
3000 if (ret == 0 && cache->io_ctl.inode) {
77745c05
JB
3001 should_put = 0;
3002
3003 /*
3004 * The cache_write_mutex is protecting the
3005 * io_list, also refer to the definition of
3006 * btrfs_transaction::io_bgs for more details
3007 */
3008 list_add_tail(&cache->io_list, io);
3009 } else {
3010 /*
3011 * If we failed to write the cache, the
3012 * generation will be bad and life goes on
3013 */
3014 ret = 0;
3015 }
3016 }
3017 if (!ret) {
3be4d8ef 3018 ret = update_block_group_item(trans, path, cache);
77745c05
JB
3019 /*
3020 * Our block group might still be attached to the list
3021 * of new block groups in the transaction handle of some
3022 * other task (struct btrfs_trans_handle->new_bgs). This
3023 * means its block group item isn't yet in the extent
3024 * tree. If this happens ignore the error, as we will
3025 * try again later in the critical section of the
3026 * transaction commit.
3027 */
3028 if (ret == -ENOENT) {
3029 ret = 0;
3030 spin_lock(&cur_trans->dirty_bgs_lock);
3031 if (list_empty(&cache->dirty_list)) {
3032 list_add_tail(&cache->dirty_list,
3033 &cur_trans->dirty_bgs);
3034 btrfs_get_block_group(cache);
3035 drop_reserve = false;
3036 }
3037 spin_unlock(&cur_trans->dirty_bgs_lock);
3038 } else if (ret) {
3039 btrfs_abort_transaction(trans, ret);
3040 }
3041 }
3042
3043 /* If it's not on the io list, we need to put the block group */
3044 if (should_put)
3045 btrfs_put_block_group(cache);
3046 if (drop_reserve)
3047 btrfs_delayed_refs_rsv_release(fs_info, 1);
77745c05
JB
3048 /*
3049 * Avoid blocking other tasks for too long. It might even save
3050 * us from writing caches for block groups that are going to be
3051 * removed.
3052 */
3053 mutex_unlock(&trans->transaction->cache_write_mutex);
938fcbfb
JB
3054 if (ret)
3055 goto out;
77745c05
JB
3056 mutex_lock(&trans->transaction->cache_write_mutex);
3057 }
3058 mutex_unlock(&trans->transaction->cache_write_mutex);
3059
3060 /*
3061 * Go through delayed refs for all the stuff we've just kicked off
3062 * and then loop back (just once)
3063 */
34d1eb0e
JB
3064 if (!ret)
3065 ret = btrfs_run_delayed_refs(trans, 0);
77745c05
JB
3066 if (!ret && loops == 0) {
3067 loops++;
3068 spin_lock(&cur_trans->dirty_bgs_lock);
3069 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3070 /*
3071 * dirty_bgs_lock protects us from concurrent block group
3072 * deletes too (not just cache_write_mutex).
3073 */
3074 if (!list_empty(&dirty)) {
3075 spin_unlock(&cur_trans->dirty_bgs_lock);
3076 goto again;
3077 }
3078 spin_unlock(&cur_trans->dirty_bgs_lock);
938fcbfb
JB
3079 }
3080out:
3081 if (ret < 0) {
3082 spin_lock(&cur_trans->dirty_bgs_lock);
3083 list_splice_init(&dirty, &cur_trans->dirty_bgs);
3084 spin_unlock(&cur_trans->dirty_bgs_lock);
77745c05
JB
3085 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3086 }
3087
3088 btrfs_free_path(path);
3089 return ret;
3090}
3091
3092int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans)
3093{
3094 struct btrfs_fs_info *fs_info = trans->fs_info;
32da5386 3095 struct btrfs_block_group *cache;
77745c05
JB
3096 struct btrfs_transaction *cur_trans = trans->transaction;
3097 int ret = 0;
3098 int should_put;
3099 struct btrfs_path *path;
3100 struct list_head *io = &cur_trans->io_bgs;
77745c05
JB
3101
3102 path = btrfs_alloc_path();
3103 if (!path)
3104 return -ENOMEM;
3105
3106 /*
3107 * Even though we are in the critical section of the transaction commit,
3108 * we can still have concurrent tasks adding elements to this
3109 * transaction's list of dirty block groups. These tasks correspond to
3110 * endio free space workers started when writeback finishes for a
3111 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3112 * allocate new block groups as a result of COWing nodes of the root
3113 * tree when updating the free space inode. The writeback for the space
3114 * caches is triggered by an earlier call to
3115 * btrfs_start_dirty_block_groups() and iterations of the following
3116 * loop.
3117 * Also we want to do the cache_save_setup first and then run the
3118 * delayed refs to make sure we have the best chance at doing this all
3119 * in one shot.
3120 */
3121 spin_lock(&cur_trans->dirty_bgs_lock);
3122 while (!list_empty(&cur_trans->dirty_bgs)) {
3123 cache = list_first_entry(&cur_trans->dirty_bgs,
32da5386 3124 struct btrfs_block_group,
77745c05
JB
3125 dirty_list);
3126
3127 /*
3128 * This can happen if cache_save_setup re-dirties a block group
3129 * that is already under IO. Just wait for it to finish and
3130 * then do it all again
3131 */
3132 if (!list_empty(&cache->io_list)) {
3133 spin_unlock(&cur_trans->dirty_bgs_lock);
3134 list_del_init(&cache->io_list);
3135 btrfs_wait_cache_io(trans, cache, path);
3136 btrfs_put_block_group(cache);
3137 spin_lock(&cur_trans->dirty_bgs_lock);
3138 }
3139
3140 /*
3141 * Don't remove from the dirty list until after we've waited on
3142 * any pending IO
3143 */
3144 list_del_init(&cache->dirty_list);
3145 spin_unlock(&cur_trans->dirty_bgs_lock);
3146 should_put = 1;
3147
3148 cache_save_setup(cache, trans, path);
3149
3150 if (!ret)
3151 ret = btrfs_run_delayed_refs(trans,
3152 (unsigned long) -1);
3153
3154 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3155 cache->io_ctl.inode = NULL;
3156 ret = btrfs_write_out_cache(trans, cache, path);
3157 if (ret == 0 && cache->io_ctl.inode) {
77745c05
JB
3158 should_put = 0;
3159 list_add_tail(&cache->io_list, io);
3160 } else {
3161 /*
3162 * If we failed to write the cache, the
3163 * generation will be bad and life goes on
3164 */
3165 ret = 0;
3166 }
3167 }
3168 if (!ret) {
3be4d8ef 3169 ret = update_block_group_item(trans, path, cache);
77745c05
JB
3170 /*
3171 * One of the free space endio workers might have
3172 * created a new block group while updating a free space
3173 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3174 * and hasn't released its transaction handle yet, in
3175 * which case the new block group is still attached to
3176 * its transaction handle and its creation has not
3177 * finished yet (no block group item in the extent tree
3178 * yet, etc). If this is the case, wait for all free
3179 * space endio workers to finish and retry. This is a
260db43c 3180 * very rare case so no need for a more efficient and
77745c05
JB
3181 * complex approach.
3182 */
3183 if (ret == -ENOENT) {
3184 wait_event(cur_trans->writer_wait,
3185 atomic_read(&cur_trans->num_writers) == 1);
3be4d8ef 3186 ret = update_block_group_item(trans, path, cache);
77745c05
JB
3187 }
3188 if (ret)
3189 btrfs_abort_transaction(trans, ret);
3190 }
3191
3192 /* If its not on the io list, we need to put the block group */
3193 if (should_put)
3194 btrfs_put_block_group(cache);
3195 btrfs_delayed_refs_rsv_release(fs_info, 1);
3196 spin_lock(&cur_trans->dirty_bgs_lock);
3197 }
3198 spin_unlock(&cur_trans->dirty_bgs_lock);
3199
3200 /*
3201 * Refer to the definition of io_bgs member for details why it's safe
3202 * to use it without any locking
3203 */
3204 while (!list_empty(io)) {
32da5386 3205 cache = list_first_entry(io, struct btrfs_block_group,
77745c05
JB
3206 io_list);
3207 list_del_init(&cache->io_list);
3208 btrfs_wait_cache_io(trans, cache, path);
3209 btrfs_put_block_group(cache);
3210 }
3211
3212 btrfs_free_path(path);
3213 return ret;
3214}
606d1bf1 3215
ac2f1e63
JB
3216static inline bool should_reclaim_block_group(struct btrfs_block_group *bg,
3217 u64 bytes_freed)
3218{
3219 const struct btrfs_space_info *space_info = bg->space_info;
3220 const int reclaim_thresh = READ_ONCE(space_info->bg_reclaim_threshold);
3221 const u64 new_val = bg->used;
3222 const u64 old_val = new_val + bytes_freed;
3223 u64 thresh;
3224
3225 if (reclaim_thresh == 0)
3226 return false;
3227
3228 thresh = div_factor_fine(bg->length, reclaim_thresh);
3229
3230 /*
3231 * If we were below the threshold before don't reclaim, we are likely a
3232 * brand new block group and we don't want to relocate new block groups.
3233 */
3234 if (old_val < thresh)
3235 return false;
3236 if (new_val >= thresh)
3237 return false;
3238 return true;
3239}
3240
606d1bf1 3241int btrfs_update_block_group(struct btrfs_trans_handle *trans,
11b66fa6 3242 u64 bytenr, u64 num_bytes, bool alloc)
606d1bf1
JB
3243{
3244 struct btrfs_fs_info *info = trans->fs_info;
32da5386 3245 struct btrfs_block_group *cache = NULL;
606d1bf1
JB
3246 u64 total = num_bytes;
3247 u64 old_val;
3248 u64 byte_in_group;
3249 int factor;
3250 int ret = 0;
3251
3252 /* Block accounting for super block */
3253 spin_lock(&info->delalloc_root_lock);
3254 old_val = btrfs_super_bytes_used(info->super_copy);
3255 if (alloc)
3256 old_val += num_bytes;
3257 else
3258 old_val -= num_bytes;
3259 btrfs_set_super_bytes_used(info->super_copy, old_val);
3260 spin_unlock(&info->delalloc_root_lock);
3261
3262 while (total) {
ac2f1e63
JB
3263 bool reclaim;
3264
606d1bf1
JB
3265 cache = btrfs_lookup_block_group(info, bytenr);
3266 if (!cache) {
3267 ret = -ENOENT;
3268 break;
3269 }
3270 factor = btrfs_bg_type_to_factor(cache->flags);
3271
3272 /*
3273 * If this block group has free space cache written out, we
3274 * need to make sure to load it if we are removing space. This
3275 * is because we need the unpinning stage to actually add the
3276 * space back to the block group, otherwise we will leak space.
3277 */
32da5386 3278 if (!alloc && !btrfs_block_group_done(cache))
ced8ecf0 3279 btrfs_cache_block_group(cache, true);
606d1bf1 3280
b3470b5d
DS
3281 byte_in_group = bytenr - cache->start;
3282 WARN_ON(byte_in_group > cache->length);
606d1bf1
JB
3283
3284 spin_lock(&cache->space_info->lock);
3285 spin_lock(&cache->lock);
3286
3287 if (btrfs_test_opt(info, SPACE_CACHE) &&
3288 cache->disk_cache_state < BTRFS_DC_CLEAR)
3289 cache->disk_cache_state = BTRFS_DC_CLEAR;
3290
bf38be65 3291 old_val = cache->used;
b3470b5d 3292 num_bytes = min(total, cache->length - byte_in_group);
606d1bf1
JB
3293 if (alloc) {
3294 old_val += num_bytes;
bf38be65 3295 cache->used = old_val;
606d1bf1
JB
3296 cache->reserved -= num_bytes;
3297 cache->space_info->bytes_reserved -= num_bytes;
3298 cache->space_info->bytes_used += num_bytes;
3299 cache->space_info->disk_used += num_bytes * factor;
3300 spin_unlock(&cache->lock);
3301 spin_unlock(&cache->space_info->lock);
3302 } else {
3303 old_val -= num_bytes;
bf38be65 3304 cache->used = old_val;
606d1bf1
JB
3305 cache->pinned += num_bytes;
3306 btrfs_space_info_update_bytes_pinned(info,
3307 cache->space_info, num_bytes);
3308 cache->space_info->bytes_used -= num_bytes;
3309 cache->space_info->disk_used -= num_bytes * factor;
ac2f1e63
JB
3310
3311 reclaim = should_reclaim_block_group(cache, num_bytes);
606d1bf1
JB
3312 spin_unlock(&cache->lock);
3313 spin_unlock(&cache->space_info->lock);
3314
fe119a6e 3315 set_extent_dirty(&trans->transaction->pinned_extents,
606d1bf1
JB
3316 bytenr, bytenr + num_bytes - 1,
3317 GFP_NOFS | __GFP_NOFAIL);
3318 }
3319
3320 spin_lock(&trans->transaction->dirty_bgs_lock);
3321 if (list_empty(&cache->dirty_list)) {
3322 list_add_tail(&cache->dirty_list,
3323 &trans->transaction->dirty_bgs);
3324 trans->delayed_ref_updates++;
3325 btrfs_get_block_group(cache);
3326 }
3327 spin_unlock(&trans->transaction->dirty_bgs_lock);
3328
3329 /*
3330 * No longer have used bytes in this block group, queue it for
3331 * deletion. We do this after adding the block group to the
3332 * dirty list to avoid races between cleaner kthread and space
3333 * cache writeout.
3334 */
6e80d4f8
DZ
3335 if (!alloc && old_val == 0) {
3336 if (!btrfs_test_opt(info, DISCARD_ASYNC))
3337 btrfs_mark_bg_unused(cache);
ac2f1e63
JB
3338 } else if (!alloc && reclaim) {
3339 btrfs_mark_bg_to_reclaim(cache);
6e80d4f8 3340 }
606d1bf1
JB
3341
3342 btrfs_put_block_group(cache);
3343 total -= num_bytes;
3344 bytenr += num_bytes;
3345 }
3346
3347 /* Modified block groups are accounted for in the delayed_refs_rsv. */
3348 btrfs_update_delayed_refs_rsv(trans);
3349 return ret;
3350}
3351
3352/**
3353 * btrfs_add_reserved_bytes - update the block_group and space info counters
3354 * @cache: The cache we are manipulating
3355 * @ram_bytes: The number of bytes of file content, and will be same to
3356 * @num_bytes except for the compress path.
3357 * @num_bytes: The number of bytes in question
3358 * @delalloc: The blocks are allocated for the delalloc write
3359 *
3360 * This is called by the allocator when it reserves space. If this is a
3361 * reservation and the block group has become read only we cannot make the
3362 * reservation and return -EAGAIN, otherwise this function always succeeds.
3363 */
32da5386 3364int btrfs_add_reserved_bytes(struct btrfs_block_group *cache,
606d1bf1
JB
3365 u64 ram_bytes, u64 num_bytes, int delalloc)
3366{
3367 struct btrfs_space_info *space_info = cache->space_info;
3368 int ret = 0;
3369
3370 spin_lock(&space_info->lock);
3371 spin_lock(&cache->lock);
3372 if (cache->ro) {
3373 ret = -EAGAIN;
3374 } else {
3375 cache->reserved += num_bytes;
3376 space_info->bytes_reserved += num_bytes;
a43c3835
JB
3377 trace_btrfs_space_reservation(cache->fs_info, "space_info",
3378 space_info->flags, num_bytes, 1);
606d1bf1
JB
3379 btrfs_space_info_update_bytes_may_use(cache->fs_info,
3380 space_info, -ram_bytes);
3381 if (delalloc)
3382 cache->delalloc_bytes += num_bytes;
99ffb43e
JB
3383
3384 /*
3385 * Compression can use less space than we reserved, so wake
3386 * tickets if that happens
3387 */
3388 if (num_bytes < ram_bytes)
3389 btrfs_try_granting_tickets(cache->fs_info, space_info);
606d1bf1
JB
3390 }
3391 spin_unlock(&cache->lock);
3392 spin_unlock(&space_info->lock);
3393 return ret;
3394}
3395
3396/**
3397 * btrfs_free_reserved_bytes - update the block_group and space info counters
3398 * @cache: The cache we are manipulating
3399 * @num_bytes: The number of bytes in question
3400 * @delalloc: The blocks are allocated for the delalloc write
3401 *
3402 * This is called by somebody who is freeing space that was never actually used
3403 * on disk. For example if you reserve some space for a new leaf in transaction
3404 * A and before transaction A commits you free that leaf, you call this with
3405 * reserve set to 0 in order to clear the reservation.
3406 */
32da5386 3407void btrfs_free_reserved_bytes(struct btrfs_block_group *cache,
606d1bf1
JB
3408 u64 num_bytes, int delalloc)
3409{
3410 struct btrfs_space_info *space_info = cache->space_info;
3411
3412 spin_lock(&space_info->lock);
3413 spin_lock(&cache->lock);
3414 if (cache->ro)
3415 space_info->bytes_readonly += num_bytes;
3416 cache->reserved -= num_bytes;
3417 space_info->bytes_reserved -= num_bytes;
3418 space_info->max_extent_size = 0;
3419
3420 if (delalloc)
3421 cache->delalloc_bytes -= num_bytes;
3422 spin_unlock(&cache->lock);
3308234a
JB
3423
3424 btrfs_try_granting_tickets(cache->fs_info, space_info);
606d1bf1
JB
3425 spin_unlock(&space_info->lock);
3426}
07730d87
JB
3427
3428static void force_metadata_allocation(struct btrfs_fs_info *info)
3429{
3430 struct list_head *head = &info->space_info;
3431 struct btrfs_space_info *found;
3432
72804905 3433 list_for_each_entry(found, head, list) {
07730d87
JB
3434 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3435 found->force_alloc = CHUNK_ALLOC_FORCE;
3436 }
07730d87
JB
3437}
3438
3439static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
3440 struct btrfs_space_info *sinfo, int force)
3441{
3442 u64 bytes_used = btrfs_space_info_used(sinfo, false);
3443 u64 thresh;
3444
3445 if (force == CHUNK_ALLOC_FORCE)
3446 return 1;
3447
3448 /*
3449 * in limited mode, we want to have some free space up to
3450 * about 1% of the FS size.
3451 */
3452 if (force == CHUNK_ALLOC_LIMITED) {
3453 thresh = btrfs_super_total_bytes(fs_info->super_copy);
3454 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
3455
3456 if (sinfo->total_bytes - bytes_used < thresh)
3457 return 1;
3458 }
3459
3460 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
3461 return 0;
3462 return 1;
3463}
3464
3465int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
3466{
3467 u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type);
3468
3469 return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
3470}
3471
820c363b 3472static struct btrfs_block_group *do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags)
79bd3712
FM
3473{
3474 struct btrfs_block_group *bg;
3475 int ret;
3476
3477 /*
3478 * Check if we have enough space in the system space info because we
3479 * will need to update device items in the chunk btree and insert a new
3480 * chunk item in the chunk btree as well. This will allocate a new
3481 * system block group if needed.
3482 */
3483 check_system_chunk(trans, flags);
3484
f6f39f7a 3485 bg = btrfs_create_chunk(trans, flags);
79bd3712
FM
3486 if (IS_ERR(bg)) {
3487 ret = PTR_ERR(bg);
3488 goto out;
3489 }
3490
79bd3712
FM
3491 ret = btrfs_chunk_alloc_add_chunk_item(trans, bg);
3492 /*
3493 * Normally we are not expected to fail with -ENOSPC here, since we have
3494 * previously reserved space in the system space_info and allocated one
ecd84d54 3495 * new system chunk if necessary. However there are three exceptions:
79bd3712
FM
3496 *
3497 * 1) We may have enough free space in the system space_info but all the
3498 * existing system block groups have a profile which can not be used
3499 * for extent allocation.
3500 *
3501 * This happens when mounting in degraded mode. For example we have a
3502 * RAID1 filesystem with 2 devices, lose one device and mount the fs
3503 * using the other device in degraded mode. If we then allocate a chunk,
3504 * we may have enough free space in the existing system space_info, but
3505 * none of the block groups can be used for extent allocation since they
3506 * have a RAID1 profile, and because we are in degraded mode with a
3507 * single device, we are forced to allocate a new system chunk with a
3508 * SINGLE profile. Making check_system_chunk() iterate over all system
3509 * block groups and check if they have a usable profile and enough space
3510 * can be slow on very large filesystems, so we tolerate the -ENOSPC and
3511 * try again after forcing allocation of a new system chunk. Like this
3512 * we avoid paying the cost of that search in normal circumstances, when
3513 * we were not mounted in degraded mode;
3514 *
3515 * 2) We had enough free space info the system space_info, and one suitable
3516 * block group to allocate from when we called check_system_chunk()
3517 * above. However right after we called it, the only system block group
3518 * with enough free space got turned into RO mode by a running scrub,
3519 * and in this case we have to allocate a new one and retry. We only
3520 * need do this allocate and retry once, since we have a transaction
ecd84d54
FM
3521 * handle and scrub uses the commit root to search for block groups;
3522 *
3523 * 3) We had one system block group with enough free space when we called
3524 * check_system_chunk(), but after that, right before we tried to
3525 * allocate the last extent buffer we needed, a discard operation came
3526 * in and it temporarily removed the last free space entry from the
3527 * block group (discard removes a free space entry, discards it, and
3528 * then adds back the entry to the block group cache).
79bd3712
FM
3529 */
3530 if (ret == -ENOSPC) {
3531 const u64 sys_flags = btrfs_system_alloc_profile(trans->fs_info);
3532 struct btrfs_block_group *sys_bg;
3533
f6f39f7a 3534 sys_bg = btrfs_create_chunk(trans, sys_flags);
79bd3712
FM
3535 if (IS_ERR(sys_bg)) {
3536 ret = PTR_ERR(sys_bg);
3537 btrfs_abort_transaction(trans, ret);
3538 goto out;
3539 }
3540
3541 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3542 if (ret) {
3543 btrfs_abort_transaction(trans, ret);
3544 goto out;
3545 }
3546
3547 ret = btrfs_chunk_alloc_add_chunk_item(trans, bg);
3548 if (ret) {
3549 btrfs_abort_transaction(trans, ret);
3550 goto out;
3551 }
3552 } else if (ret) {
3553 btrfs_abort_transaction(trans, ret);
3554 goto out;
3555 }
3556out:
3557 btrfs_trans_release_chunk_metadata(trans);
3558
820c363b
NA
3559 if (ret)
3560 return ERR_PTR(ret);
3561
3562 btrfs_get_block_group(bg);
3563 return bg;
79bd3712
FM
3564}
3565
07730d87 3566/*
79bd3712
FM
3567 * Chunk allocation is done in 2 phases:
3568 *
3569 * 1) Phase 1 - through btrfs_chunk_alloc() we allocate device extents for
3570 * the chunk, the chunk mapping, create its block group and add the items
3571 * that belong in the chunk btree to it - more specifically, we need to
3572 * update device items in the chunk btree and add a new chunk item to it.
3573 *
3574 * 2) Phase 2 - through btrfs_create_pending_block_groups(), we add the block
3575 * group item to the extent btree and the device extent items to the devices
3576 * btree.
3577 *
3578 * This is done to prevent deadlocks. For example when COWing a node from the
3579 * extent btree we are holding a write lock on the node's parent and if we
3580 * trigger chunk allocation and attempted to insert the new block group item
3581 * in the extent btree right way, we could deadlock because the path for the
3582 * insertion can include that parent node. At first glance it seems impossible
3583 * to trigger chunk allocation after starting a transaction since tasks should
3584 * reserve enough transaction units (metadata space), however while that is true
3585 * most of the time, chunk allocation may still be triggered for several reasons:
3586 *
3587 * 1) When reserving metadata, we check if there is enough free space in the
3588 * metadata space_info and therefore don't trigger allocation of a new chunk.
3589 * However later when the task actually tries to COW an extent buffer from
3590 * the extent btree or from the device btree for example, it is forced to
3591 * allocate a new block group (chunk) because the only one that had enough
3592 * free space was just turned to RO mode by a running scrub for example (or
3593 * device replace, block group reclaim thread, etc), so we can not use it
3594 * for allocating an extent and end up being forced to allocate a new one;
3595 *
3596 * 2) Because we only check that the metadata space_info has enough free bytes,
3597 * we end up not allocating a new metadata chunk in that case. However if
3598 * the filesystem was mounted in degraded mode, none of the existing block
3599 * groups might be suitable for extent allocation due to their incompatible
3600 * profile (for e.g. mounting a 2 devices filesystem, where all block groups
3601 * use a RAID1 profile, in degraded mode using a single device). In this case
3602 * when the task attempts to COW some extent buffer of the extent btree for
3603 * example, it will trigger allocation of a new metadata block group with a
3604 * suitable profile (SINGLE profile in the example of the degraded mount of
3605 * the RAID1 filesystem);
3606 *
3607 * 3) The task has reserved enough transaction units / metadata space, but when
3608 * it attempts to COW an extent buffer from the extent or device btree for
3609 * example, it does not find any free extent in any metadata block group,
3610 * therefore forced to try to allocate a new metadata block group.
3611 * This is because some other task allocated all available extents in the
3612 * meanwhile - this typically happens with tasks that don't reserve space
3613 * properly, either intentionally or as a bug. One example where this is
3614 * done intentionally is fsync, as it does not reserve any transaction units
3615 * and ends up allocating a variable number of metadata extents for log
ecd84d54
FM
3616 * tree extent buffers;
3617 *
3618 * 4) The task has reserved enough transaction units / metadata space, but right
3619 * before it tries to allocate the last extent buffer it needs, a discard
3620 * operation comes in and, temporarily, removes the last free space entry from
3621 * the only metadata block group that had free space (discard starts by
3622 * removing a free space entry from a block group, then does the discard
3623 * operation and, once it's done, it adds back the free space entry to the
3624 * block group).
79bd3712
FM
3625 *
3626 * We also need this 2 phases setup when adding a device to a filesystem with
3627 * a seed device - we must create new metadata and system chunks without adding
3628 * any of the block group items to the chunk, extent and device btrees. If we
3629 * did not do it this way, we would get ENOSPC when attempting to update those
3630 * btrees, since all the chunks from the seed device are read-only.
3631 *
3632 * Phase 1 does the updates and insertions to the chunk btree because if we had
3633 * it done in phase 2 and have a thundering herd of tasks allocating chunks in
3634 * parallel, we risk having too many system chunks allocated by many tasks if
3635 * many tasks reach phase 1 without the previous ones completing phase 2. In the
3636 * extreme case this leads to exhaustion of the system chunk array in the
3637 * superblock. This is easier to trigger if using a btree node/leaf size of 64K
3638 * and with RAID filesystems (so we have more device items in the chunk btree).
3639 * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of
3640 * the system chunk array due to concurrent allocations") provides more details.
3641 *
2bb2e00e
FM
3642 * Allocation of system chunks does not happen through this function. A task that
3643 * needs to update the chunk btree (the only btree that uses system chunks), must
3644 * preallocate chunk space by calling either check_system_chunk() or
3645 * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or
3646 * metadata chunk or when removing a chunk, while the later is used before doing
3647 * a modification to the chunk btree - use cases for the later are adding,
3648 * removing and resizing a device as well as relocation of a system chunk.
3649 * See the comment below for more details.
79bd3712
FM
3650 *
3651 * The reservation of system space, done through check_system_chunk(), as well
3652 * as all the updates and insertions into the chunk btree must be done while
3653 * holding fs_info->chunk_mutex. This is important to guarantee that while COWing
3654 * an extent buffer from the chunks btree we never trigger allocation of a new
3655 * system chunk, which would result in a deadlock (trying to lock twice an
3656 * extent buffer of the chunk btree, first time before triggering the chunk
3657 * allocation and the second time during chunk allocation while attempting to
3658 * update the chunks btree). The system chunk array is also updated while holding
3659 * that mutex. The same logic applies to removing chunks - we must reserve system
3660 * space, update the chunk btree and the system chunk array in the superblock
3661 * while holding fs_info->chunk_mutex.
3662 *
3663 * This function, btrfs_chunk_alloc(), belongs to phase 1.
3664 *
3665 * If @force is CHUNK_ALLOC_FORCE:
07730d87
JB
3666 * - return 1 if it successfully allocates a chunk,
3667 * - return errors including -ENOSPC otherwise.
79bd3712 3668 * If @force is NOT CHUNK_ALLOC_FORCE:
07730d87
JB
3669 * - return 0 if it doesn't need to allocate a new chunk,
3670 * - return 1 if it successfully allocates a chunk,
3671 * - return errors including -ENOSPC otherwise.
3672 */
3673int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
3674 enum btrfs_chunk_alloc_enum force)
3675{
3676 struct btrfs_fs_info *fs_info = trans->fs_info;
3677 struct btrfs_space_info *space_info;
820c363b 3678 struct btrfs_block_group *ret_bg;
07730d87
JB
3679 bool wait_for_alloc = false;
3680 bool should_alloc = false;
760e69c4 3681 bool from_extent_allocation = false;
07730d87
JB
3682 int ret = 0;
3683
760e69c4
NA
3684 if (force == CHUNK_ALLOC_FORCE_FOR_EXTENT) {
3685 from_extent_allocation = true;
3686 force = CHUNK_ALLOC_FORCE;
3687 }
3688
07730d87
JB
3689 /* Don't re-enter if we're already allocating a chunk */
3690 if (trans->allocating_chunk)
3691 return -ENOSPC;
79bd3712 3692 /*
2bb2e00e
FM
3693 * Allocation of system chunks can not happen through this path, as we
3694 * could end up in a deadlock if we are allocating a data or metadata
3695 * chunk and there is another task modifying the chunk btree.
3696 *
3697 * This is because while we are holding the chunk mutex, we will attempt
3698 * to add the new chunk item to the chunk btree or update an existing
3699 * device item in the chunk btree, while the other task that is modifying
3700 * the chunk btree is attempting to COW an extent buffer while holding a
3701 * lock on it and on its parent - if the COW operation triggers a system
3702 * chunk allocation, then we can deadlock because we are holding the
3703 * chunk mutex and we may need to access that extent buffer or its parent
3704 * in order to add the chunk item or update a device item.
3705 *
3706 * Tasks that want to modify the chunk tree should reserve system space
3707 * before updating the chunk btree, by calling either
3708 * btrfs_reserve_chunk_metadata() or check_system_chunk().
3709 * It's possible that after a task reserves the space, it still ends up
3710 * here - this happens in the cases described above at do_chunk_alloc().
3711 * The task will have to either retry or fail.
79bd3712 3712 */
2bb2e00e 3713 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
79bd3712 3714 return -ENOSPC;
07730d87
JB
3715
3716 space_info = btrfs_find_space_info(fs_info, flags);
3717 ASSERT(space_info);
3718
3719 do {
3720 spin_lock(&space_info->lock);
3721 if (force < space_info->force_alloc)
3722 force = space_info->force_alloc;
3723 should_alloc = should_alloc_chunk(fs_info, space_info, force);
3724 if (space_info->full) {
3725 /* No more free physical space */
3726 if (should_alloc)
3727 ret = -ENOSPC;
3728 else
3729 ret = 0;
3730 spin_unlock(&space_info->lock);
3731 return ret;
3732 } else if (!should_alloc) {
3733 spin_unlock(&space_info->lock);
3734 return 0;
3735 } else if (space_info->chunk_alloc) {
3736 /*
3737 * Someone is already allocating, so we need to block
3738 * until this someone is finished and then loop to
3739 * recheck if we should continue with our allocation
3740 * attempt.
3741 */
3742 wait_for_alloc = true;
1314ca78 3743 force = CHUNK_ALLOC_NO_FORCE;
07730d87
JB
3744 spin_unlock(&space_info->lock);
3745 mutex_lock(&fs_info->chunk_mutex);
3746 mutex_unlock(&fs_info->chunk_mutex);
3747 } else {
3748 /* Proceed with allocation */
3749 space_info->chunk_alloc = 1;
3750 wait_for_alloc = false;
3751 spin_unlock(&space_info->lock);
3752 }
3753
3754 cond_resched();
3755 } while (wait_for_alloc);
3756
3757 mutex_lock(&fs_info->chunk_mutex);
3758 trans->allocating_chunk = true;
3759
3760 /*
3761 * If we have mixed data/metadata chunks we want to make sure we keep
3762 * allocating mixed chunks instead of individual chunks.
3763 */
3764 if (btrfs_mixed_space_info(space_info))
3765 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3766
3767 /*
3768 * if we're doing a data chunk, go ahead and make sure that
3769 * we keep a reasonable number of metadata chunks allocated in the
3770 * FS as well.
3771 */
3772 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3773 fs_info->data_chunk_allocations++;
3774 if (!(fs_info->data_chunk_allocations %
3775 fs_info->metadata_ratio))
3776 force_metadata_allocation(fs_info);
3777 }
3778
820c363b 3779 ret_bg = do_chunk_alloc(trans, flags);
07730d87
JB
3780 trans->allocating_chunk = false;
3781
760e69c4 3782 if (IS_ERR(ret_bg)) {
820c363b 3783 ret = PTR_ERR(ret_bg);
760e69c4
NA
3784 } else if (from_extent_allocation) {
3785 /*
3786 * New block group is likely to be used soon. Try to activate
3787 * it now. Failure is OK for now.
3788 */
3789 btrfs_zone_activate(ret_bg);
3790 }
3791
3792 if (!ret)
820c363b
NA
3793 btrfs_put_block_group(ret_bg);
3794
07730d87
JB
3795 spin_lock(&space_info->lock);
3796 if (ret < 0) {
3797 if (ret == -ENOSPC)
3798 space_info->full = 1;
3799 else
3800 goto out;
3801 } else {
3802 ret = 1;
3803 space_info->max_extent_size = 0;
3804 }
3805
3806 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3807out:
3808 space_info->chunk_alloc = 0;
3809 spin_unlock(&space_info->lock);
3810 mutex_unlock(&fs_info->chunk_mutex);
07730d87
JB
3811
3812 return ret;
3813}
3814
3815static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
3816{
3817 u64 num_dev;
3818
3819 num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max;
3820 if (!num_dev)
3821 num_dev = fs_info->fs_devices->rw_devices;
3822
3823 return num_dev;
3824}
3825
2bb2e00e
FM
3826static void reserve_chunk_space(struct btrfs_trans_handle *trans,
3827 u64 bytes,
3828 u64 type)
07730d87
JB
3829{
3830 struct btrfs_fs_info *fs_info = trans->fs_info;
3831 struct btrfs_space_info *info;
3832 u64 left;
07730d87 3833 int ret = 0;
07730d87
JB
3834
3835 /*
3836 * Needed because we can end up allocating a system chunk and for an
3837 * atomic and race free space reservation in the chunk block reserve.
3838 */
3839 lockdep_assert_held(&fs_info->chunk_mutex);
3840
3841 info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3842 spin_lock(&info->lock);
3843 left = info->total_bytes - btrfs_space_info_used(info, true);
3844 spin_unlock(&info->lock);
3845
2bb2e00e 3846 if (left < bytes && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
07730d87 3847 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
2bb2e00e 3848 left, bytes, type);
07730d87
JB
3849 btrfs_dump_space_info(fs_info, info, 0, 0);
3850 }
3851
2bb2e00e 3852 if (left < bytes) {
07730d87 3853 u64 flags = btrfs_system_alloc_profile(fs_info);
79bd3712 3854 struct btrfs_block_group *bg;
07730d87
JB
3855
3856 /*
3857 * Ignore failure to create system chunk. We might end up not
3858 * needing it, as we might not need to COW all nodes/leafs from
3859 * the paths we visit in the chunk tree (they were already COWed
3860 * or created in the current transaction for example).
3861 */
f6f39f7a 3862 bg = btrfs_create_chunk(trans, flags);
79bd3712
FM
3863 if (IS_ERR(bg)) {
3864 ret = PTR_ERR(bg);
2bb2e00e 3865 } else {
b6a98021
NA
3866 /*
3867 * We have a new chunk. We also need to activate it for
3868 * zoned filesystem.
3869 */
3870 ret = btrfs_zoned_activate_one_bg(fs_info, info, true);
3871 if (ret < 0)
3872 return;
3873
79bd3712
FM
3874 /*
3875 * If we fail to add the chunk item here, we end up
3876 * trying again at phase 2 of chunk allocation, at
3877 * btrfs_create_pending_block_groups(). So ignore
2bb2e00e
FM
3878 * any error here. An ENOSPC here could happen, due to
3879 * the cases described at do_chunk_alloc() - the system
3880 * block group we just created was just turned into RO
3881 * mode by a scrub for example, or a running discard
3882 * temporarily removed its free space entries, etc.
79bd3712
FM
3883 */
3884 btrfs_chunk_alloc_add_chunk_item(trans, bg);
3885 }
07730d87
JB
3886 }
3887
3888 if (!ret) {
9270501c 3889 ret = btrfs_block_rsv_add(fs_info,
07730d87 3890 &fs_info->chunk_block_rsv,
2bb2e00e 3891 bytes, BTRFS_RESERVE_NO_FLUSH);
1cb3db1c 3892 if (!ret)
2bb2e00e 3893 trans->chunk_bytes_reserved += bytes;
07730d87
JB
3894 }
3895}
3896
2bb2e00e
FM
3897/*
3898 * Reserve space in the system space for allocating or removing a chunk.
3899 * The caller must be holding fs_info->chunk_mutex.
3900 */
3901void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
3902{
3903 struct btrfs_fs_info *fs_info = trans->fs_info;
3904 const u64 num_devs = get_profile_num_devs(fs_info, type);
3905 u64 bytes;
3906
3907 /* num_devs device items to update and 1 chunk item to add or remove. */
3908 bytes = btrfs_calc_metadata_size(fs_info, num_devs) +
3909 btrfs_calc_insert_metadata_size(fs_info, 1);
3910
3911 reserve_chunk_space(trans, bytes, type);
3912}
3913
3914/*
3915 * Reserve space in the system space, if needed, for doing a modification to the
3916 * chunk btree.
3917 *
3918 * @trans: A transaction handle.
3919 * @is_item_insertion: Indicate if the modification is for inserting a new item
3920 * in the chunk btree or if it's for the deletion or update
3921 * of an existing item.
3922 *
3923 * This is used in a context where we need to update the chunk btree outside
3924 * block group allocation and removal, to avoid a deadlock with a concurrent
3925 * task that is allocating a metadata or data block group and therefore needs to
3926 * update the chunk btree while holding the chunk mutex. After the update to the
3927 * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called.
3928 *
3929 */
3930void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans,
3931 bool is_item_insertion)
3932{
3933 struct btrfs_fs_info *fs_info = trans->fs_info;
3934 u64 bytes;
3935
3936 if (is_item_insertion)
3937 bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
3938 else
3939 bytes = btrfs_calc_metadata_size(fs_info, 1);
3940
3941 mutex_lock(&fs_info->chunk_mutex);
3942 reserve_chunk_space(trans, bytes, BTRFS_BLOCK_GROUP_SYSTEM);
3943 mutex_unlock(&fs_info->chunk_mutex);
3944}
3945
3e43c279
JB
3946void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
3947{
32da5386 3948 struct btrfs_block_group *block_group;
3e43c279
JB
3949 u64 last = 0;
3950
3951 while (1) {
3952 struct inode *inode;
3953
3954 block_group = btrfs_lookup_first_block_group(info, last);
3955 while (block_group) {
3956 btrfs_wait_block_group_cache_done(block_group);
3957 spin_lock(&block_group->lock);
3958 if (block_group->iref)
3959 break;
3960 spin_unlock(&block_group->lock);
3961 block_group = btrfs_next_block_group(block_group);
3962 }
3963 if (!block_group) {
3964 if (last == 0)
3965 break;
3966 last = 0;
3967 continue;
3968 }
3969
3970 inode = block_group->inode;
3971 block_group->iref = 0;
3972 block_group->inode = NULL;
3973 spin_unlock(&block_group->lock);
3974 ASSERT(block_group->io_ctl.inode == NULL);
3975 iput(inode);
b3470b5d 3976 last = block_group->start + block_group->length;
3e43c279
JB
3977 btrfs_put_block_group(block_group);
3978 }
3979}
3980
3981/*
3982 * Must be called only after stopping all workers, since we could have block
3983 * group caching kthreads running, and therefore they could race with us if we
3984 * freed the block groups before stopping them.
3985 */
3986int btrfs_free_block_groups(struct btrfs_fs_info *info)
3987{
32da5386 3988 struct btrfs_block_group *block_group;
3e43c279
JB
3989 struct btrfs_space_info *space_info;
3990 struct btrfs_caching_control *caching_ctl;
3991 struct rb_node *n;
3992
16b0c258 3993 write_lock(&info->block_group_cache_lock);
3e43c279
JB
3994 while (!list_empty(&info->caching_block_groups)) {
3995 caching_ctl = list_entry(info->caching_block_groups.next,
3996 struct btrfs_caching_control, list);
3997 list_del(&caching_ctl->list);
3998 btrfs_put_caching_control(caching_ctl);
3999 }
16b0c258 4000 write_unlock(&info->block_group_cache_lock);
3e43c279
JB
4001
4002 spin_lock(&info->unused_bgs_lock);
4003 while (!list_empty(&info->unused_bgs)) {
4004 block_group = list_first_entry(&info->unused_bgs,
32da5386 4005 struct btrfs_block_group,
3e43c279
JB
4006 bg_list);
4007 list_del_init(&block_group->bg_list);
4008 btrfs_put_block_group(block_group);
4009 }
3e43c279 4010
18bb8bbf
JT
4011 while (!list_empty(&info->reclaim_bgs)) {
4012 block_group = list_first_entry(&info->reclaim_bgs,
4013 struct btrfs_block_group,
4014 bg_list);
4015 list_del_init(&block_group->bg_list);
4016 btrfs_put_block_group(block_group);
4017 }
4018 spin_unlock(&info->unused_bgs_lock);
4019
afba2bc0
NA
4020 spin_lock(&info->zone_active_bgs_lock);
4021 while (!list_empty(&info->zone_active_bgs)) {
4022 block_group = list_first_entry(&info->zone_active_bgs,
4023 struct btrfs_block_group,
4024 active_bg_list);
4025 list_del_init(&block_group->active_bg_list);
4026 btrfs_put_block_group(block_group);
4027 }
4028 spin_unlock(&info->zone_active_bgs_lock);
4029
16b0c258 4030 write_lock(&info->block_group_cache_lock);
08dddb29 4031 while ((n = rb_last(&info->block_group_cache_tree.rb_root)) != NULL) {
32da5386 4032 block_group = rb_entry(n, struct btrfs_block_group,
3e43c279 4033 cache_node);
08dddb29
FM
4034 rb_erase_cached(&block_group->cache_node,
4035 &info->block_group_cache_tree);
3e43c279 4036 RB_CLEAR_NODE(&block_group->cache_node);
16b0c258 4037 write_unlock(&info->block_group_cache_lock);
3e43c279
JB
4038
4039 down_write(&block_group->space_info->groups_sem);
4040 list_del(&block_group->list);
4041 up_write(&block_group->space_info->groups_sem);
4042
4043 /*
4044 * We haven't cached this block group, which means we could
4045 * possibly have excluded extents on this block group.
4046 */
4047 if (block_group->cached == BTRFS_CACHE_NO ||
4048 block_group->cached == BTRFS_CACHE_ERROR)
4049 btrfs_free_excluded_extents(block_group);
4050
4051 btrfs_remove_free_space_cache(block_group);
4052 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
4053 ASSERT(list_empty(&block_group->dirty_list));
4054 ASSERT(list_empty(&block_group->io_list));
4055 ASSERT(list_empty(&block_group->bg_list));
48aaeebe 4056 ASSERT(refcount_read(&block_group->refs) == 1);
195a49ea 4057 ASSERT(block_group->swap_extents == 0);
3e43c279
JB
4058 btrfs_put_block_group(block_group);
4059
16b0c258 4060 write_lock(&info->block_group_cache_lock);
3e43c279 4061 }
16b0c258 4062 write_unlock(&info->block_group_cache_lock);
3e43c279 4063
3e43c279
JB
4064 btrfs_release_global_block_rsv(info);
4065
4066 while (!list_empty(&info->space_info)) {
4067 space_info = list_entry(info->space_info.next,
4068 struct btrfs_space_info,
4069 list);
4070
4071 /*
4072 * Do not hide this behind enospc_debug, this is actually
4073 * important and indicates a real bug if this happens.
4074 */
4075 if (WARN_ON(space_info->bytes_pinned > 0 ||
3e43c279
JB
4076 space_info->bytes_may_use > 0))
4077 btrfs_dump_space_info(info, space_info, 0, 0);
40cdc509
FM
4078
4079 /*
4080 * If there was a failure to cleanup a log tree, very likely due
4081 * to an IO failure on a writeback attempt of one or more of its
4082 * extent buffers, we could not do proper (and cheap) unaccounting
4083 * of their reserved space, so don't warn on bytes_reserved > 0 in
4084 * that case.
4085 */
4086 if (!(space_info->flags & BTRFS_BLOCK_GROUP_METADATA) ||
4087 !BTRFS_FS_LOG_CLEANUP_ERROR(info)) {
4088 if (WARN_ON(space_info->bytes_reserved > 0))
4089 btrfs_dump_space_info(info, space_info, 0, 0);
4090 }
4091
d611add4 4092 WARN_ON(space_info->reclaim_size > 0);
3e43c279
JB
4093 list_del(&space_info->list);
4094 btrfs_sysfs_remove_space_info(space_info);
4095 }
4096 return 0;
4097}
684b752b
FM
4098
4099void btrfs_freeze_block_group(struct btrfs_block_group *cache)
4100{
4101 atomic_inc(&cache->frozen);
4102}
4103
4104void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group)
4105{
4106 struct btrfs_fs_info *fs_info = block_group->fs_info;
4107 struct extent_map_tree *em_tree;
4108 struct extent_map *em;
4109 bool cleanup;
4110
4111 spin_lock(&block_group->lock);
4112 cleanup = (atomic_dec_and_test(&block_group->frozen) &&
4113 block_group->removed);
4114 spin_unlock(&block_group->lock);
4115
4116 if (cleanup) {
684b752b
FM
4117 em_tree = &fs_info->mapping_tree;
4118 write_lock(&em_tree->lock);
4119 em = lookup_extent_mapping(em_tree, block_group->start,
4120 1);
4121 BUG_ON(!em); /* logic error, can't happen */
4122 remove_extent_mapping(em_tree, em);
4123 write_unlock(&em_tree->lock);
684b752b
FM
4124
4125 /* once for us and once for the tree */
4126 free_extent_map(em);
4127 free_extent_map(em);
4128
4129 /*
4130 * We may have left one free space entry and other possible
4131 * tasks trimming this block group have left 1 entry each one.
4132 * Free them if any.
4133 */
4134 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
4135 }
4136}
195a49ea
FM
4137
4138bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg)
4139{
4140 bool ret = true;
4141
4142 spin_lock(&bg->lock);
4143 if (bg->ro)
4144 ret = false;
4145 else
4146 bg->swap_extents++;
4147 spin_unlock(&bg->lock);
4148
4149 return ret;
4150}
4151
4152void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount)
4153{
4154 spin_lock(&bg->lock);
4155 ASSERT(!bg->ro);
4156 ASSERT(bg->swap_extents >= amount);
4157 bg->swap_extents -= amount;
4158 spin_unlock(&bg->lock);
4159}